bool:\n- return len(seq) > len(set(seq))\n-\n-\n-@pytest.mark.parametrize(\n- 'framework,expected',\n- [\n- ('qt', 'qtagg'),\n- ('gtk3', 'gtk3agg'),\n- ('gtk4', 'gtk4agg'),\n- ('wx', 'wxagg'),\n- ('tk', 'tkagg'),\n- ('macosx', 'macosx'),\n- ('headless', 'agg'),\n- ('does not exist', None),\n- ]\n-)\n-def test_backend_for_gui_framework(framework, expected):\n- assert backend_registry.backend_for_gui_framework(framework) == expected\n-\n-\n-def test_list_builtin():\n- backends = backend_registry.list_builtin()\n- assert not has_duplicates(backends)\n- # Compare using sets as order is not important\n- assert {*backends} == {\n- 'gtk3agg', 'gtk3cairo', 'gtk4agg', 'gtk4cairo', 'macosx', 'nbagg', 'notebook',\n- 'qtagg', 'qtcairo', 'qt5agg', 'qt5cairo', 'tkagg',\n- 'tkcairo', 'webagg', 'wx', 'wxagg', 'wxcairo', 'agg', 'cairo', 'pdf', 'pgf',\n- 'ps', 'svg', 'template',\n- }\n-\n-\n-@pytest.mark.parametrize(\n- 'filter,expected',\n- [\n- (BackendFilter.INTERACTIVE,\n- ['gtk3agg', 'gtk3cairo', 'gtk4agg', 'gtk4cairo', 'macosx', 'nbagg', 'notebook',\n- 'qtagg', 'qtcairo', 'qt5agg', 'qt5cairo', 'tkagg',\n- 'tkcairo', 'webagg', 'wx', 'wxagg', 'wxcairo']),\n- (BackendFilter.NON_INTERACTIVE,\n- ['agg', 'cairo', 'pdf', 'pgf', 'ps', 'svg', 'template']),\n- ]\n-)\n-def test_list_builtin_with_filter(filter, expected):\n- backends = backend_registry.list_builtin(filter)\n- assert not has_duplicates(backends)\n- # Compare using sets as order is not important\n- assert {*backends} == {*expected}\n-\n-\n-def test_list_gui_frameworks():\n- frameworks = backend_registry.list_gui_frameworks()\n- assert not has_duplicates(frameworks)\n- # Compare using sets as order is not important\n- assert {*frameworks} == {\n- \"gtk3\", \"gtk4\", \"macosx\", \"qt\", \"qt5\", \"qt6\", \"tk\", \"wx\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"backend, is_valid\", [\n- (\"agg\", True),\n- (\"QtAgg\", True),\n- (\"module://anything\", True),\n- (\"made-up-name\", False),\n-])\n-def test_is_valid_backend(backend, is_valid):\n- assert backend_registry.is_valid_backend(backend) == is_valid\n-\n-\n-@pytest.mark.parametrize(\"backend, normalized\", [\n- (\"agg\", \"matplotlib.backends.backend_agg\"),\n- (\"QtAgg\", \"matplotlib.backends.backend_qtagg\"),\n- (\"module://Anything\", \"Anything\"),\n-])\n-def test_backend_normalization(backend, normalized):\n- assert backend_registry._backend_module_name(backend) == normalized\n-\n-\n-def test_entry_points_inline():\n- pytest.importorskip('matplotlib_inline')\n- backends = backend_registry.list_all()\n- assert 'inline' in backends\n-\n-\n-def test_entry_points_ipympl():\n- pytest.importorskip('ipympl')\n- backends = backend_registry.list_all()\n- assert 'ipympl' in backends\n- assert 'widget' in backends\n-\n-\n-def test_entry_point_name_shadows_builtin(clear_backend_registry):\n- with pytest.raises(RuntimeError):\n- backend_registry._validate_and_store_entry_points(\n- [('qtagg', 'module1')])\n-\n-\n-def test_entry_point_name_duplicate(clear_backend_registry):\n- with pytest.raises(RuntimeError):\n- backend_registry._validate_and_store_entry_points(\n- [('some_name', 'module1'), ('some_name', 'module2')])\n-\n-\n-def test_entry_point_identical(clear_backend_registry):\n- # Issue https://github.com/matplotlib/matplotlib/issues/28367\n- # Multiple entry points with the same name and value (value is the module)\n- # are acceptable.\n- n = len(backend_registry._name_to_module)\n- backend_registry._validate_and_store_entry_points(\n- [('some_name', 'some.module'), ('some_name', 'some.module')])\n- assert len(backend_registry._name_to_module) == n+1\n- assert backend_registry._name_to_module['some_name'] == 'module://some.module'\n-\n-\n-def test_entry_point_name_is_module(clear_backend_registry):\n- with pytest.raises(RuntimeError):\n- backend_registry._validate_and_store_entry_points(\n- [('module://backend.something', 'module1')])\n-\n-\n-@pytest.mark.parametrize('backend', [\n- 'agg',\n- 'module://matplotlib.backends.backend_agg',\n-])\n-def test_load_entry_points_only_if_needed(clear_backend_registry, backend):\n- assert not backend_registry._loaded_entry_points\n- check = backend_registry.resolve_backend(backend)\n- assert check == (backend, None)\n- assert not backend_registry._loaded_entry_points\n- backend_registry.list_all() # Force load of entry points\n- assert backend_registry._loaded_entry_points\n-\n-\n-@pytest.mark.parametrize(\n- 'gui_or_backend, expected_backend, expected_gui',\n- [\n- ('agg', 'agg', None),\n- ('qt', 'qtagg', 'qt'),\n- ('TkCairo', 'tkcairo', 'tk'),\n- ]\n-)\n-def test_resolve_gui_or_backend(gui_or_backend, expected_backend, expected_gui):\n- backend, gui = backend_registry.resolve_gui_or_backend(gui_or_backend)\n- assert backend == expected_backend\n- assert gui == expected_gui\n-\n-\n-def test_resolve_gui_or_backend_invalid():\n- match = \"is not a recognised GUI loop or backend name\"\n- with pytest.raises(RuntimeError, match=match):\n- backend_registry.resolve_gui_or_backend('no-such-name')\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S37\",\n \"shared_functions\": 1,\n \"shared_lines\": 4\n },\n {\n \"spec_id\": \"S38\",\n \"shared_functions\": 1,\n \"shared_lines\": 4\n },\n {\n \"spec_id\": \"S40\",\n \"shared_functions\": 1,\n \"shared_lines\": 4\n }\n ]\n}"}}},{"rowIdx":55,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S1\",\n \"text\": \"Create a new webhook. This method creates a webhook that will be triggered when specified MLflow events occur. Webhooks allow you to receive HTTP notifications when events like model version creation, transition, or deletion happen in your MLflow deployment. Signature: create_webhook(self, name: str, url: str, events: list[WebhookEventStr | WebhookEvent], description: str | None = None, secret: str | None = None, status: str | WebhookStatus | None = None) -> Webhook. Args: name: Name for the webhook. This should be a descriptive identifier for the webhook. url: Webhook endpoint URL where HTTP POST requests will be sent when events occur. The URL must be accessible from the MLflow server. description: Optional description of the webhook's purpose or functionality. secret: Optional secret string used for HMAC-SHA256 signature verification. If provided, MLflow will include an X-MLflow-Signature header in webhook requests that can be used to verify the request authenticity. Returns: A Webhook object representing the created webhook, containing the webhook ID, configuration details, and metadata like creation timestamp. Task statement: Create, retrieve, update, and delete webhooks for MLflow model registry events; Support webhook registration with configurable URLs, event types, and optional secrets.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracking/client.py\",\n \"lines\": \"6310-6355\",\n \"function\": \"MlflowClient.create_webhook\"\n },\n {\n \"file\": \"mlflow/tracking/_model_registry/client.py\",\n \"lines\": \"798-823\",\n \"function\": \"ModelRegistryClient.create_webhook\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n--- a/mlflow/entities/webhook.py\n+++ b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n- def __eq__(self, other: object) -> bool:\n- if not isinstance(other, WebhookEvent):\n- return False\n- return self.entity == other.entity and self.action == other.action\n+\n+\n+\n+\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n- @property\n- def secret(self) -> str | None:\n- return self._secret\n+\n+\n+\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n--- a/mlflow/server/handlers.py\n+++ b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n--- a/mlflow/store/artifact/azure_blob_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-from mlflow.utils.credentials import get_default_host_creds\n+\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- augmented_raise_for_status,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n- def __init__(self, store_uri=None, tracking_uri=None):\n- \"\"\"\n- Empty constructor. This is deliberately not marked as abstract, else every derived class\n- would be forced to create one.\n \n- Args:\n- store_uri: The model registry store URI.\n- tracking_uri: URI of the current MLflow tracking server, used to perform operations\n- like fetching source run metadata or downloading source run artifacts\n- to support subsequently uploading them to the model registry storage\n- location.\n- \"\"\"\n- # Create a thread lock to ensure thread safety when linking prompts to other entities,\n- # since the default linking implementation reads and appends entity tags, which\n- # is prone to concurrent modification issues\n- self._prompt_link_lock = threading.RLock()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n--- a/mlflow/store/model_registry/base_rest_store.py\n+++ b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n-from mlflow.utils.rest_utils import (\n- call_endpoint,\n- call_endpoints,\n-)\n+from mlflow.utils.rest_utils import call_endpoints\n+\n+\n+\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- @record_usage_event(CreateWebhookEvent)\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Unique name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of event types that trigger this webhook.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE).\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- return self.store.create_webhook(name, url, events, description, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self.store.get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self.store.list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of event types.\n- secret: New webhook secret.\n- status: New webhook status.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self.store.delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n- return self.store.test_webhook(webhook_id, event)\n+ return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n--- a/mlflow/tracking/_model_registry/registry.py\n+++ b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n- from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n+\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n--- a/mlflow/tracking/_model_registry/utils.py\n+++ b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n-def _get_registry_uri_from_spark_session():\n- session = _get_active_spark_session()\n- if session is None:\n- return None\n \n- if is_in_databricks_serverless_runtime():\n- # Connected to Serverless\n- return \"databricks-uc\"\n \n- from pyspark.sql.utils import AnalysisException\n \n- try:\n- return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n- except AnalysisException:\n- # In serverless clusters, session.conf.get() is unsupported\n- # and raises an AnalysisException. We may encounter this case\n- # when DBConnect is used to connect to a serverless cluster,\n- # in which case the prior `is_in_databricks_serverless_runtime()`\n- # check will have returned false (as of 2025-06-07, it checks\n- # an environment variable that isn't set by DBConnect)\n- return None\n \n \n-def _get_registry_uri_from_context():\n- if _registry_uri is not None:\n- return _registry_uri\n- elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n- return uri\n- return _registry_uri\n \n \n-def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n- \"\"\"\n- Get the default registry URI for a given tracking URI.\n \n- If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n- Otherwise, returns the tracking URI itself.\n \n- Args:\n- tracking_uri: The tracking URI to get the default registry URI for\n \n- Returns:\n- The default registry URI\n- \"\"\"\n- if tracking_uri is not None and is_databricks_uri(tracking_uri):\n- # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n- # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n- # model registry offering on Databricks\n- if tracking_uri == \"databricks\":\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n- else:\n- # Extract profile from tracking URI and construct databricks-uc URI\n- profile, key_prefix = get_db_info_from_uri(tracking_uri)\n- if profile:\n- # Reconstruct the profile string including key_prefix if present\n- profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n- return construct_db_uc_uri_from_profile(profile_string)\n- else:\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n-\n- # For non-databricks tracking URIs, use the tracking URI as the registry URI\n- return tracking_uri\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n-def _resolve_registry_uri(\n- registry_uri: str | None = None, tracking_uri: str | None = None\n-) -> str | None:\n- \"\"\"\n- Resolve the registry URI following the same logic as get_registry_uri().\n- \"\"\"\n- return (\n- registry_uri\n- or _get_registry_uri_from_context()\n- or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n-def _get_rest_store(store_uri, **_):\n- return RestStore(partial(get_default_host_creds, store_uri))\n+\n+\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n-def _get_store_registry():\n- global _model_registry_store_registry\n- from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n- from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n- if _model_registry_store_registry is not None:\n- return _model_registry_store_registry\n \n- _model_registry_store_registry = ModelRegistryStoreRegistry()\n- _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n- # Register a placeholder function that raises if users pass a registry URI with scheme\n- # \"databricks-uc\"\n- _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n- _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n- for scheme in [\"http\", \"https\"]:\n- _model_registry_store_registry.register(scheme, _get_rest_store)\n \n- for scheme in DATABASE_ENGINES:\n- _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n- for scheme in [\"\", \"file\"]:\n- _model_registry_store_registry.register(scheme, _get_file_store)\n \n- _model_registry_store_registry.register_entrypoints()\n- return _model_registry_store_registry\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n--- a/mlflow/tracking/_tracking_service/utils.py\n+++ b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n- @property\n- def tracking_uri(self):\n- return self._tracking_client.tracking_uri\n \n- def _get_registry_client(self):\n- \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n- Raises:\n- MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n- example, when the registry URI refers to an unsupported store type (e.g., the\n- FileStore).\n \n- Returns:\n- A ModelRegistryClient instance.\n- \"\"\"\n- # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n- # an instance variable on this `MlflowClient` instance. Because the instance variable\n- # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n- # function is used to safely fetch the variable (if it is defined) or a NoneType\n- # (if it is not defined)\n- registry_client_attr = \"_registry_client_lazy\"\n- registry_client = getattr(self, registry_client_attr, None)\n- if registry_client is None:\n- try:\n- registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n- # Define an instance variable on this `MlflowClient` instance to reference the\n- # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n- # that the variable name is consistent with the variable name specified in the\n- # preceding call to `getattr()`\n- setattr(self, registry_client_attr, registry_client)\n- except UnsupportedModelRegistryStoreURIException as exc:\n- raise MlflowException(\n- \"Model Registry features are not supported by the store with URI:\"\n- f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n- f\" {exc.supported_uri_schemes}.\",\n- FEATURE_DISABLED,\n- )\n- return registry_client\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEventStr | WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of events that trigger this webhook. Can be strings or\n- `WebhookEvent` objects.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self._get_registry_client().get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self._get_registry_client().list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEventStr | WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of events. Can be strings or `WebhookEvent` objects.\n- secret: New webhook secret.\n- status: New webhook status. Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().update_webhook(\n- webhook_id=webhook_id,\n- name=name,\n- description=description,\n- url=url,\n- events=events,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self._get_registry_client().delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n- return self._get_registry_client().test_webhook(webhook_id, event)\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,7 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,36 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n--- a/mlflow/utils/credentials.py\n+++ b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n-def get_default_host_creds(store_uri):\n- creds = read_mlflow_creds()\n- return MlflowHostCreds(\n- host=store_uri,\n- username=creds.username,\n- password=creds.password,\n- token=MLFLOW_TRACKING_TOKEN.get(),\n- aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n- auth=MLFLOW_TRACKING_AUTH.get(),\n- ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n- client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n- server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n-def parse_dict(js_dict, message):\n- \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n- _stringify_all_experiment_ids(js_dict)\n- ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n+\n+\n+\n+\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n--- a/mlflow/utils/request_utils.py\n+++ b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n-@lru_cache(maxsize=64)\n-def _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- # To create a new Session object for each process, we use the process id as the cache key.\n- # This is to avoid sharing the same Session object across processes, which can lead to issues\n- # such as https://stackoverflow.com/q/3724900.\n- _pid,\n- respect_retry_after_header=True,\n-):\n- \"\"\"\n- This function should not be called directly. Instead, use `_get_request_session` below.\n- \"\"\"\n \n- retry_kwargs = {\n- \"total\": max_retries,\n- \"connect\": max_retries,\n- \"read\": max_retries,\n- \"redirect\": max_retries,\n- \"status\": max_retries,\n- \"status_forcelist\": retry_codes,\n- \"backoff_factor\": backoff_factor,\n- \"backoff_jitter\": backoff_jitter,\n- \"raise_on_status\": raise_on_status,\n- \"respect_retry_after_header\": respect_retry_after_header,\n- }\n- urllib3_version = Version(urllib3.__version__)\n- if urllib3_version >= Version(\"1.26.0\"):\n- retry_kwargs[\"allowed_methods\"] = None\n- else:\n- retry_kwargs[\"method_whitelist\"] = None\n-\n- if urllib3_version < Version(\"2.0\"):\n- retry = JitteredRetry(**retry_kwargs)\n- else:\n- retry = Retry(**retry_kwargs)\n- from mlflow.environment_variables import (\n- MLFLOW_HTTP_POOL_CONNECTIONS,\n- MLFLOW_HTTP_POOL_MAXSIZE,\n- )\n-\n- adapter = HTTPAdapter(\n- pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n- pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n- max_retries=retry,\n- )\n- session = requests.Session()\n- session.mount(\"https://\", adapter)\n- session.mount(\"http://\", adapter)\n- return session\n-\n-\n-def _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n-):\n- \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n- Args:\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n- Returns:\n- requests.Session object.\n \n- \"\"\"\n- return _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- _pid=os.getpid(),\n- respect_retry_after_header=respect_retry_after_header,\n- )\n \n \n-def _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status=True,\n- allow_redirects=None,\n- respect_retry_after_header=True,\n- **kwargs,\n-):\n- \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n- Args:\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- url: The target URL address for the HTTP request.\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- session = _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n- )\n \n- # the environment variable is hardcoded here to avoid importing mlflow.\n- # however, documentation is available in environment_variables.py\n- env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n- allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n- return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n-from mlflow.utils.proto_json_utils import parse_dict\n-from mlflow.utils.request_utils import (\n- _TRANSIENT_FAILURE_RESPONSE_CODES,\n- _get_http_response_with_retries,\n- augmented_raise_for_status, # noqa: F401\n- cloud_storage_http_request, # noqa: F401\n-)\n-from mlflow.utils.string_utils import strip_suffix\n+\n+from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n+\n+\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n \n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n \n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n-\n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n \n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n-\n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n-\n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n-\n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n-\n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n-\n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n-\n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n-def _can_parse_as_json_object(string):\n- try:\n- return isinstance(json.loads(string), dict)\n- except Exception:\n- return False\n+\n+\n+\n+\n+\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n-def verify_rest_response(response, endpoint):\n- \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n- # Handle Armeria-specific response case where response text is \"200 OK\"\n- # v1/traces endpoint might return empty response\n- if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n- response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n- return response\n-\n- # Handle non-200 status codes\n- if response.status_code != 200:\n- if _can_parse_as_json_object(response.text):\n- raise RestException(json.loads(response.text))\n- else:\n- base_msg = (\n- f\"API request to endpoint {endpoint} \"\n- f\"failed with error code {response.status_code} != 200\"\n- )\n- raise MlflowException(\n- f\"{base_msg}. Response body: '{response.text}'\",\n- error_code=get_error_code(response.status_code),\n- )\n \n- # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n- # response\n- if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n- base_msg = (\n- \"API request to endpoint was successful but the response body was not \"\n- \"in a valid JSON format\"\n- )\n- raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n-\n- return response\n-\n-\n-def _validate_max_retries(max_retries):\n- max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n-\n- if max_retry_limit < 0:\n- raise MlflowException(\n- message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n- \"Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if max_retries >= max_retry_limit:\n- raise MlflowException(\n- message=f\"The configured max_retries value ({max_retries}) is \"\n- f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if max_retries < 0:\n- raise MlflowException(\n- message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n-\n-def _validate_backoff_factor(backoff_factor):\n- max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n-\n- if max_backoff_factor_limit < 0:\n- raise MlflowException(\n- message=\"The current maximum backoff factor limit is invalid \"\n- f\"({max_backoff_factor_limit}). Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor >= max_backoff_factor_limit:\n- raise MlflowException(\n- message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n- \"of the maximum allowable backoff_factor limit \"\n- f\"({max_backoff_factor_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor < 0:\n- raise MlflowException(\n- message=\"The backoff_factor value must be either 0 a positive integer. \"\n- f\"Got {backoff_factor}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n-def call_endpoint(\n- host_creds,\n- endpoint,\n- method,\n- json_body,\n- response_proto,\n- extra_headers=None,\n- retry_timeout_seconds=None,\n-):\n- # Convert json string to json dictionary, to pass to requests\n- if json_body is not None:\n- json_body = json.loads(json_body)\n- call_kwargs = {\n- \"host_creds\": host_creds,\n- \"endpoint\": endpoint,\n- \"method\": method,\n- }\n- if extra_headers is not None:\n- call_kwargs[\"extra_headers\"] = extra_headers\n- if retry_timeout_seconds is not None:\n- call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n- if method == \"GET\":\n- call_kwargs[\"params\"] = json_body\n- response = http_request(**call_kwargs)\n- else:\n- call_kwargs[\"json\"] = json_body\n- response = http_request(**call_kwargs)\n \n- response = verify_rest_response(response, endpoint)\n- response_to_parse = response.text\n- try:\n- js_dict = json.loads(response_to_parse)\n- except json.JSONDecodeError:\n- if len(response_to_parse) > 50:\n- response_to_parse = response_to_parse[:50] + \"...\"\n- _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n- raise\n \n- parse_dict(js_dict=js_dict, message=response_proto)\n- return response_proto\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n- def __init__(\n- self,\n- host,\n- username=None,\n- password=None,\n- token=None,\n- aws_sigv4=False,\n- auth=None,\n- ignore_tls_verification=False,\n- client_cert_path=None,\n- server_cert_path=None,\n- use_databricks_sdk=False,\n- databricks_auth_profile=None,\n- client_id=None,\n- client_secret=None,\n- use_secret_scope_token=False,\n- ):\n- if not host:\n- raise MlflowException(\n- message=\"host is a required parameter for MlflowHostCreds\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if ignore_tls_verification and (server_cert_path is not None):\n- raise MlflowException(\n- message=(\n- \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n- \"must not be set! This error may have occurred because the \"\n- \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n- \"environment variables are both set - only one of these environment \"\n- \"variables may be set.\"\n- ),\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- self.host = host\n- self.username = username\n- self.password = password\n- self.token = token\n- self.aws_sigv4 = aws_sigv4\n- self.auth = auth\n- self.ignore_tls_verification = ignore_tls_verification\n- self.client_cert_path = client_cert_path\n- self.server_cert_path = server_cert_path\n- self.use_databricks_sdk = use_databricks_sdk\n- self.databricks_auth_profile = databricks_auth_profile\n- self.client_id = client_id\n- self.client_secret = client_secret\n- self.use_secret_scope_token = use_secret_scope_token\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,12 +712,11 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n- @property\n- def verify(self):\n- if self.use_databricks_sdk:\n- # Let databricks-sdk set HTTP request `verify` param.\n- return None\n- if self.server_cert_path is None:\n- return not self.ignore_tls_verification\n- else:\n- return self.server_cert_path\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n+++ a/mlflow/entities/webhook.py\n--- b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n+ def __eq__(self, other: object) -> bool:\n+ if not isinstance(other, WebhookEvent):\n+ return False\n+ return self.entity == other.entity and self.action == other.action\n-\n-\n-\n-\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n+ @property\n+ def secret(self) -> str | None:\n+ return self._secret\n-\n-\n-\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n+++ a/mlflow/server/handlers.py\n--- b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n+++ a/mlflow/store/artifact/azure_blob_artifact_repo.py\n--- b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+from mlflow.utils.credentials import get_default_host_creds\n-\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ augmented_raise_for_status,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(self, store_uri=None, tracking_uri=None):\n+ \"\"\"\n+ Empty constructor. This is deliberately not marked as abstract, else every derived class\n+ would be forced to create one.\n \n+ Args:\n+ store_uri: The model registry store URI.\n+ tracking_uri: URI of the current MLflow tracking server, used to perform operations\n+ like fetching source run metadata or downloading source run artifacts\n+ to support subsequently uploading them to the model registry storage\n+ location.\n+ \"\"\"\n+ # Create a thread lock to ensure thread safety when linking prompts to other entities,\n+ # since the default linking implementation reads and appends entity tags, which\n+ # is prone to concurrent modification issues\n+ self._prompt_link_lock = threading.RLock()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n+++ a/mlflow/store/model_registry/base_rest_store.py\n--- b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n+from mlflow.utils.rest_utils import (\n+ call_endpoint,\n+ call_endpoints,\n+)\n-from mlflow.utils.rest_utils import call_endpoints\n-\n-\n-\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n+++ a/mlflow/tracking/_model_registry/client.py\n--- b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ @record_usage_event(CreateWebhookEvent)\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Unique name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of event types that trigger this webhook.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE).\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ return self.store.create_webhook(name, url, events, description, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self.store.get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self.store.list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of event types.\n+ secret: New webhook secret.\n+ status: New webhook status.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self.store.delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n+ return self.store.test_webhook(webhook_id, event)\n- return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n+++ a/mlflow/tracking/_model_registry/registry.py\n--- b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n+ from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n-\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n+++ a/mlflow/tracking/_model_registry/utils.py\n--- b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n+def _get_registry_uri_from_spark_session():\n+ session = _get_active_spark_session()\n+ if session is None:\n+ return None\n \n+ if is_in_databricks_serverless_runtime():\n+ # Connected to Serverless\n+ return \"databricks-uc\"\n \n+ from pyspark.sql.utils import AnalysisException\n \n+ try:\n+ return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n+ except AnalysisException:\n+ # In serverless clusters, session.conf.get() is unsupported\n+ # and raises an AnalysisException. We may encounter this case\n+ # when DBConnect is used to connect to a serverless cluster,\n+ # in which case the prior `is_in_databricks_serverless_runtime()`\n+ # check will have returned false (as of 2025-06-07, it checks\n+ # an environment variable that isn't set by DBConnect)\n+ return None\n \n \n+def _get_registry_uri_from_context():\n+ if _registry_uri is not None:\n+ return _registry_uri\n+ elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n+ return uri\n+ return _registry_uri\n \n \n+def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n+ \"\"\"\n+ Get the default registry URI for a given tracking URI.\n \n+ If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n+ Otherwise, returns the tracking URI itself.\n \n+ Args:\n+ tracking_uri: The tracking URI to get the default registry URI for\n \n+ Returns:\n+ The default registry URI\n+ \"\"\"\n+ if tracking_uri is not None and is_databricks_uri(tracking_uri):\n+ # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n+ # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n+ # model registry offering on Databricks\n+ if tracking_uri == \"databricks\":\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+ else:\n+ # Extract profile from tracking URI and construct databricks-uc URI\n+ profile, key_prefix = get_db_info_from_uri(tracking_uri)\n+ if profile:\n+ # Reconstruct the profile string including key_prefix if present\n+ profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n+ return construct_db_uc_uri_from_profile(profile_string)\n+ else:\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+\n+ # For non-databricks tracking URIs, use the tracking URI as the registry URI\n+ return tracking_uri\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n+def _resolve_registry_uri(\n+ registry_uri: str | None = None, tracking_uri: str | None = None\n+) -> str | None:\n+ \"\"\"\n+ Resolve the registry URI following the same logic as get_registry_uri().\n+ \"\"\"\n+ return (\n+ registry_uri\n+ or _get_registry_uri_from_context()\n+ or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n+def _get_rest_store(store_uri, **_):\n+ return RestStore(partial(get_default_host_creds, store_uri))\n-\n-\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n+def _get_store_registry():\n+ global _model_registry_store_registry\n+ from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n+ from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n+ if _model_registry_store_registry is not None:\n+ return _model_registry_store_registry\n \n+ _model_registry_store_registry = ModelRegistryStoreRegistry()\n+ _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n+ # Register a placeholder function that raises if users pass a registry URI with scheme\n+ # \"databricks-uc\"\n+ _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n+ _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n+ for scheme in [\"http\", \"https\"]:\n+ _model_registry_store_registry.register(scheme, _get_rest_store)\n \n+ for scheme in DATABASE_ENGINES:\n+ _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n+ for scheme in [\"\", \"file\"]:\n+ _model_registry_store_registry.register(scheme, _get_file_store)\n \n+ _model_registry_store_registry.register_entrypoints()\n+ return _model_registry_store_registry\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n+++ a/mlflow/tracking/_tracking_service/utils.py\n--- b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n+++ a/mlflow/tracking/client.py\n--- b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n+ @property\n+ def tracking_uri(self):\n+ return self._tracking_client.tracking_uri\n \n+ def _get_registry_client(self):\n+ \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n+ Raises:\n+ MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n+ example, when the registry URI refers to an unsupported store type (e.g., the\n+ FileStore).\n \n+ Returns:\n+ A ModelRegistryClient instance.\n+ \"\"\"\n+ # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n+ # an instance variable on this `MlflowClient` instance. Because the instance variable\n+ # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n+ # function is used to safely fetch the variable (if it is defined) or a NoneType\n+ # (if it is not defined)\n+ registry_client_attr = \"_registry_client_lazy\"\n+ registry_client = getattr(self, registry_client_attr, None)\n+ if registry_client is None:\n+ try:\n+ registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n+ # Define an instance variable on this `MlflowClient` instance to reference the\n+ # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n+ # that the variable name is consistent with the variable name specified in the\n+ # preceding call to `getattr()`\n+ setattr(self, registry_client_attr, registry_client)\n+ except UnsupportedModelRegistryStoreURIException as exc:\n+ raise MlflowException(\n+ \"Model Registry features are not supported by the store with URI:\"\n+ f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n+ f\" {exc.supported_uri_schemes}.\",\n+ FEATURE_DISABLED,\n+ )\n+ return registry_client\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEventStr | WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of events that trigger this webhook. Can be strings or\n+ `WebhookEvent` objects.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self._get_registry_client().get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self._get_registry_client().list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEventStr | WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of events. Can be strings or `WebhookEvent` objects.\n+ secret: New webhook secret.\n+ status: New webhook status. Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ if events is not None:\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().update_webhook(\n+ webhook_id=webhook_id,\n+ name=name,\n+ description=description,\n+ url=url,\n+ events=events,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self._get_registry_client().delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n- return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,7 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,36 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n+++ a/mlflow/utils/credentials.py\n--- b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n+def get_default_host_creds(store_uri):\n+ creds = read_mlflow_creds()\n+ return MlflowHostCreds(\n+ host=store_uri,\n+ username=creds.username,\n+ password=creds.password,\n+ token=MLFLOW_TRACKING_TOKEN.get(),\n+ aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n+ auth=MLFLOW_TRACKING_AUTH.get(),\n+ ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n+ client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n+ server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n+def parse_dict(js_dict, message):\n+ \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n+ _stringify_all_experiment_ids(js_dict)\n+ ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n-\n-\n-\n-\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n+++ a/mlflow/utils/request_utils.py\n--- b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n+@lru_cache(maxsize=64)\n+def _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ # To create a new Session object for each process, we use the process id as the cache key.\n+ # This is to avoid sharing the same Session object across processes, which can lead to issues\n+ # such as https://stackoverflow.com/q/3724900.\n+ _pid,\n+ respect_retry_after_header=True,\n+):\n+ \"\"\"\n+ This function should not be called directly. Instead, use `_get_request_session` below.\n+ \"\"\"\n \n+ retry_kwargs = {\n+ \"total\": max_retries,\n+ \"connect\": max_retries,\n+ \"read\": max_retries,\n+ \"redirect\": max_retries,\n+ \"status\": max_retries,\n+ \"status_forcelist\": retry_codes,\n+ \"backoff_factor\": backoff_factor,\n+ \"backoff_jitter\": backoff_jitter,\n+ \"raise_on_status\": raise_on_status,\n+ \"respect_retry_after_header\": respect_retry_after_header,\n+ }\n+ urllib3_version = Version(urllib3.__version__)\n+ if urllib3_version >= Version(\"1.26.0\"):\n+ retry_kwargs[\"allowed_methods\"] = None\n+ else:\n+ retry_kwargs[\"method_whitelist\"] = None\n+\n+ if urllib3_version < Version(\"2.0\"):\n+ retry = JitteredRetry(**retry_kwargs)\n+ else:\n+ retry = Retry(**retry_kwargs)\n+ from mlflow.environment_variables import (\n+ MLFLOW_HTTP_POOL_CONNECTIONS,\n+ MLFLOW_HTTP_POOL_MAXSIZE,\n+ )\n+\n+ adapter = HTTPAdapter(\n+ pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n+ pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n+ max_retries=retry,\n+ )\n+ session = requests.Session()\n+ session.mount(\"https://\", adapter)\n+ session.mount(\"http://\", adapter)\n+ return session\n+\n+\n+def _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+):\n+ \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n+ Args:\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n+ Returns:\n+ requests.Session object.\n \n+ \"\"\"\n+ return _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ _pid=os.getpid(),\n+ respect_retry_after_header=respect_retry_after_header,\n+ )\n \n \n+def _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status=True,\n+ allow_redirects=None,\n+ respect_retry_after_header=True,\n+ **kwargs,\n+):\n+ \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n+ Args:\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ url: The target URL address for the HTTP request.\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ session = _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+ )\n \n+ # the environment variable is hardcoded here to avoid importing mlflow.\n+ # however, documentation is available in environment_variables.py\n+ env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n+ allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n+ return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n+from mlflow.utils.proto_json_utils import parse_dict\n+from mlflow.utils.request_utils import (\n+ _TRANSIENT_FAILURE_RESPONSE_CODES,\n+ _get_http_response_with_retries,\n+ augmented_raise_for_status, # noqa: F401\n+ cloud_storage_http_request, # noqa: F401\n+)\n+from mlflow.utils.string_utils import strip_suffix\n-\n-from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n-\n-\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n \n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n \n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n+\n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n \n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n+\n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n+\n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n+\n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n+\n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n+\n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n+\n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n+def _can_parse_as_json_object(string):\n+ try:\n+ return isinstance(json.loads(string), dict)\n+ except Exception:\n+ return False\n-\n-\n-\n-\n-\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n+def verify_rest_response(response, endpoint):\n+ \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n+ # Handle Armeria-specific response case where response text is \"200 OK\"\n+ # v1/traces endpoint might return empty response\n+ if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n+ response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n+ return response\n+\n+ # Handle non-200 status codes\n+ if response.status_code != 200:\n+ if _can_parse_as_json_object(response.text):\n+ raise RestException(json.loads(response.text))\n+ else:\n+ base_msg = (\n+ f\"API request to endpoint {endpoint} \"\n+ f\"failed with error code {response.status_code} != 200\"\n+ )\n+ raise MlflowException(\n+ f\"{base_msg}. Response body: '{response.text}'\",\n+ error_code=get_error_code(response.status_code),\n+ )\n \n+ # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n+ # response\n+ if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n+ base_msg = (\n+ \"API request to endpoint was successful but the response body was not \"\n+ \"in a valid JSON format\"\n+ )\n+ raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n+\n+ return response\n+\n+\n+def _validate_max_retries(max_retries):\n+ max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n+\n+ if max_retry_limit < 0:\n+ raise MlflowException(\n+ message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n+ \"Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if max_retries >= max_retry_limit:\n+ raise MlflowException(\n+ message=f\"The configured max_retries value ({max_retries}) is \"\n+ f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if max_retries < 0:\n+ raise MlflowException(\n+ message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+\n+def _validate_backoff_factor(backoff_factor):\n+ max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n+\n+ if max_backoff_factor_limit < 0:\n+ raise MlflowException(\n+ message=\"The current maximum backoff factor limit is invalid \"\n+ f\"({max_backoff_factor_limit}). Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor >= max_backoff_factor_limit:\n+ raise MlflowException(\n+ message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n+ \"of the maximum allowable backoff_factor limit \"\n+ f\"({max_backoff_factor_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor < 0:\n+ raise MlflowException(\n+ message=\"The backoff_factor value must be either 0 a positive integer. \"\n+ f\"Got {backoff_factor}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n+def call_endpoint(\n+ host_creds,\n+ endpoint,\n+ method,\n+ json_body,\n+ response_proto,\n+ extra_headers=None,\n+ retry_timeout_seconds=None,\n+):\n+ # Convert json string to json dictionary, to pass to requests\n+ if json_body is not None:\n+ json_body = json.loads(json_body)\n+ call_kwargs = {\n+ \"host_creds\": host_creds,\n+ \"endpoint\": endpoint,\n+ \"method\": method,\n+ }\n+ if extra_headers is not None:\n+ call_kwargs[\"extra_headers\"] = extra_headers\n+ if retry_timeout_seconds is not None:\n+ call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n+ if method == \"GET\":\n+ call_kwargs[\"params\"] = json_body\n+ response = http_request(**call_kwargs)\n+ else:\n+ call_kwargs[\"json\"] = json_body\n+ response = http_request(**call_kwargs)\n \n+ response = verify_rest_response(response, endpoint)\n+ response_to_parse = response.text\n+ try:\n+ js_dict = json.loads(response_to_parse)\n+ except json.JSONDecodeError:\n+ if len(response_to_parse) > 50:\n+ response_to_parse = response_to_parse[:50] + \"...\"\n+ _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n+ raise\n \n+ parse_dict(js_dict=js_dict, message=response_proto)\n+ return response_proto\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n+ def __init__(\n+ self,\n+ host,\n+ username=None,\n+ password=None,\n+ token=None,\n+ aws_sigv4=False,\n+ auth=None,\n+ ignore_tls_verification=False,\n+ client_cert_path=None,\n+ server_cert_path=None,\n+ use_databricks_sdk=False,\n+ databricks_auth_profile=None,\n+ client_id=None,\n+ client_secret=None,\n+ use_secret_scope_token=False,\n+ ):\n+ if not host:\n+ raise MlflowException(\n+ message=\"host is a required parameter for MlflowHostCreds\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if ignore_tls_verification and (server_cert_path is not None):\n+ raise MlflowException(\n+ message=(\n+ \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n+ \"must not be set! This error may have occurred because the \"\n+ \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n+ \"environment variables are both set - only one of these environment \"\n+ \"variables may be set.\"\n+ ),\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ self.host = host\n+ self.username = username\n+ self.password = password\n+ self.token = token\n+ self.aws_sigv4 = aws_sigv4\n+ self.auth = auth\n+ self.ignore_tls_verification = ignore_tls_verification\n+ self.client_cert_path = client_cert_path\n+ self.server_cert_path = server_cert_path\n+ self.use_databricks_sdk = use_databricks_sdk\n+ self.databricks_auth_profile = databricks_auth_profile\n+ self.client_id = client_id\n+ self.client_secret = client_secret\n+ self.use_secret_scope_token = use_secret_scope_token\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,11 +712,12 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n+ @property\n+ def verify(self):\n+ if self.use_databricks_sdk:\n+ # Let databricks-sdk set HTTP request `verify` param.\n+ return None\n+ if self.server_cert_path is None:\n+ return not self.ignore_tls_verification\n+ else:\n+ return self.server_cert_path\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..203c145da 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -6337,6 +6337,9 @@ class MlflowClient:\n if status is not None:\n status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ if description is not None:\n+ name = f\"{name}-webhook\"\n+\n return self._get_registry_client().create_webhook(\n name=name,\n url=url,\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py\ndeleted file mode 100644\nindex 97fd61b00..000000000\n--- a/tests/tracking/test_client_webhooks.py\n+++ /dev/null\n@@ -1,219 +0,0 @@\n-from pathlib import Path\n-from typing import Iterator\n-\n-import pytest\n-from cryptography.fernet import Fernet\n-\n-from mlflow.entities.webhook import WebhookAction, WebhookEntity, WebhookEvent, WebhookStatus\n-from mlflow.environment_variables import MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY\n-from mlflow.exceptions import MlflowException\n-from mlflow.server import handlers\n-from mlflow.server.fastapi_app import app\n-from mlflow.server.handlers import initialize_backend_stores\n-from mlflow.tracking import MlflowClient\n-\n-from tests.helper_functions import get_safe_port\n-from tests.tracking.integration_test_utils import ServerThread\n-\n-\n-@pytest.fixture\n-def client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Iterator[MlflowClient]:\n- \"\"\"Setup a local MLflow server with proper webhook encryption key support.\"\"\"\n- # Set up encryption key for webhooks using monkeypatch\n- encryption_key = Fernet.generate_key().decode(\"utf-8\")\n- monkeypatch.setenv(MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY.name, encryption_key)\n-\n- # Configure backend stores\n- backend_uri = f\"sqlite:///{tmp_path / 'mlflow.db'}\"\n- default_artifact_root = tmp_path.as_uri()\n-\n- # Force-reset backend stores before each test\n- handlers._tracking_store = None\n- handlers._model_registry_store = None\n- initialize_backend_stores(backend_uri, default_artifact_root=default_artifact_root)\n-\n- # Start server and return client\n- with ServerThread(app, get_safe_port()) as url:\n- yield MlflowClient(url)\n-\n-\n-def test_create_webhook(client: MlflowClient):\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n- assert webhook.events == [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n-\n- webhook = client.get_webhook(webhook.webhook_id)\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n-\n- # With secret\n- webhook_with_secret = client.create_webhook(\n- name=\"test_webhook_with_secret\",\n- url=\"https://example.com/webhook_with_secret\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- secret=\"my_secret\",\n- )\n- assert webhook_with_secret.name == \"test_webhook_with_secret\"\n- assert webhook_with_secret.url == \"https://example.com/webhook_with_secret\"\n- assert webhook_with_secret.secret is None\n- assert webhook_with_secret.events == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)\n- ]\n-\n- # Multiple events\n- webhook_multiple_events = client.create_webhook(\n- name=\"test_webhook_multiple_events\",\n- url=\"https://example.com/webhook_multiple_events\",\n- events=[\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- ],\n- )\n- assert webhook_multiple_events.name == \"test_webhook_multiple_events\"\n- assert webhook_multiple_events.url == \"https://example.com/webhook_multiple_events\"\n- assert sorted(\n- webhook_multiple_events.events, key=lambda e: (e.entity.value, e.action.value)\n- ) == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- ]\n- assert webhook_multiple_events.secret is None\n-\n-\n-def test_get_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- created_webhook = client.create_webhook(\n- name=\"test_webhook\", url=\"https://example.com/webhook\", events=events\n- )\n- retrieved_webhook = client.get_webhook(created_webhook.webhook_id)\n- assert retrieved_webhook.webhook_id == created_webhook.webhook_id\n- assert retrieved_webhook.name == \"test_webhook\"\n- assert retrieved_webhook.url == \"https://example.com/webhook\"\n- assert retrieved_webhook.events == events\n-\n-\n-def test_get_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.get_webhook(\"nonexistent\")\n-\n-\n-def test_list_webhooks(client: MlflowClient):\n- # Create more webhooks than max_results\n- created_webhooks = []\n- for i in range(5):\n- webhook = client.create_webhook(\n- name=f\"webhook{i}\",\n- url=f\"https://example.com/{i}\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- created_webhooks.append(webhook)\n- # Test pagination with max_results=2\n- webhooks_page = client.list_webhooks(max_results=2)\n- assert len(webhooks_page) == 2\n- assert webhooks_page.token is not None\n- # Get next page\n- next_webhooks_page = client.list_webhooks(max_results=2, page_token=webhooks_page.token)\n- assert len(next_webhooks_page) == 2\n- assert next_webhooks_page.token is not None\n- # Verify we don't get duplicates\n- first_page_ids = {w.webhook_id for w in webhooks_page}\n- second_page_ids = {w.webhook_id for w in next_webhooks_page}\n- assert first_page_ids.isdisjoint(second_page_ids)\n-\n-\n-def test_update_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\", url=\"https://example.com/original\", events=events\n- )\n- # Update webhook\n- new_events = [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.REGISTERED_MODEL, WebhookAction.CREATED),\n- ]\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- url=\"https://example.com/updated\",\n- events=new_events,\n- description=\"Updated description\",\n- secret=\"new_secret\",\n- status=WebhookStatus.DISABLED,\n- )\n- assert updated_webhook.webhook_id == webhook.webhook_id\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/updated\"\n- assert updated_webhook.events == new_events\n- assert updated_webhook.description == \"Updated description\"\n- assert updated_webhook.status == WebhookStatus.DISABLED\n- assert updated_webhook.last_updated_timestamp > webhook.last_updated_timestamp\n-\n-\n-def test_update_webhook_partial(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\",\n- url=\"https://example.com/original\",\n- events=events,\n- description=\"Original description\",\n- )\n- # Update only the name\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- )\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/original\"\n- assert updated_webhook.events == events\n- assert updated_webhook.description == \"Original description\"\n-\n-\n-def test_update_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.update_webhook(webhook_id=\"nonexistent\", name=\"new_name\")\n-\n-\n-@pytest.mark.parametrize(\n- (\"invalid_url\", \"expected_match\"),\n- [\n- (\" \", r\"Webhook URL cannot be empty or just whitespace\"),\n- (\"ftp://example.com\", r\"Invalid webhook URL scheme\"),\n- (\"http://[invalid\", r\"Invalid webhook URL\"),\n- ],\n-)\n-def test_update_webhook_invalid_urls(client: MlflowClient, invalid_url: str, expected_match: str):\n- # Create a valid webhook first\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- with pytest.raises(MlflowException, match=expected_match):\n- client.update_webhook(webhook_id=webhook.webhook_id, url=invalid_url)\n-\n-\n-def test_delete_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=events,\n- )\n- client.delete_webhook(webhook.webhook_id)\n- with pytest.raises(MlflowException, match=r\"Webhook with ID .* not found\"):\n- client.get_webhook(webhook.webhook_id)\n- webhooks_page = client.list_webhooks()\n- webhook_ids = {w.webhook_id for w in webhooks_page}\n- assert webhook.webhook_id not in webhook_ids\n-\n-\n-def test_delete_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.delete_webhook(\"nonexistent\")\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S3\",\n \"shared_functions\": 2,\n \"shared_lines\": 28\n },\n {\n \"spec_id\": \"S4\",\n \"shared_functions\": 2,\n \"shared_lines\": 5\n },\n {\n \"spec_id\": \"S2\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S13\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S14\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S5\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S6\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S7\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":56,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S4\",\n \"text\": \"Raises: MlflowException: If webhook creation fails due to invalid parameters, network issues, or server-side errors. Common causes include invalid URL format, unsupported event types, or insufficient permissions. Task statement: Implement robust error handling for webhook delivery failures; Ensure proper validation of webhook event types and entity-action combinations.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracking/client.py\",\n \"lines\": \"6352-6355\",\n \"function\": \"MlflowClient.create_webhook\"\n },\n {\n \"file\": \"mlflow/tracking/_model_registry/client.py\",\n \"lines\": \"823\",\n \"function\": \"ModelRegistryClient.create_webhook\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n--- a/mlflow/entities/webhook.py\n+++ b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n- def __eq__(self, other: object) -> bool:\n- if not isinstance(other, WebhookEvent):\n- return False\n- return self.entity == other.entity and self.action == other.action\n+\n+\n+\n+\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n- @property\n- def secret(self) -> str | None:\n- return self._secret\n+\n+\n+\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n--- a/mlflow/server/handlers.py\n+++ b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n--- a/mlflow/store/artifact/azure_blob_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-from mlflow.utils.credentials import get_default_host_creds\n+\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- augmented_raise_for_status,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n- def __init__(self, store_uri=None, tracking_uri=None):\n- \"\"\"\n- Empty constructor. This is deliberately not marked as abstract, else every derived class\n- would be forced to create one.\n \n- Args:\n- store_uri: The model registry store URI.\n- tracking_uri: URI of the current MLflow tracking server, used to perform operations\n- like fetching source run metadata or downloading source run artifacts\n- to support subsequently uploading them to the model registry storage\n- location.\n- \"\"\"\n- # Create a thread lock to ensure thread safety when linking prompts to other entities,\n- # since the default linking implementation reads and appends entity tags, which\n- # is prone to concurrent modification issues\n- self._prompt_link_lock = threading.RLock()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n--- a/mlflow/store/model_registry/base_rest_store.py\n+++ b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n-from mlflow.utils.rest_utils import (\n- call_endpoint,\n- call_endpoints,\n-)\n+from mlflow.utils.rest_utils import call_endpoints\n+\n+\n+\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- @record_usage_event(CreateWebhookEvent)\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Unique name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of event types that trigger this webhook.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE).\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- return self.store.create_webhook(name, url, events, description, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self.store.get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self.store.list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of event types.\n- secret: New webhook secret.\n- status: New webhook status.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self.store.delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n- return self.store.test_webhook(webhook_id, event)\n+ return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n--- a/mlflow/tracking/_model_registry/registry.py\n+++ b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n- from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n+\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n--- a/mlflow/tracking/_model_registry/utils.py\n+++ b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n-def _get_registry_uri_from_spark_session():\n- session = _get_active_spark_session()\n- if session is None:\n- return None\n \n- if is_in_databricks_serverless_runtime():\n- # Connected to Serverless\n- return \"databricks-uc\"\n \n- from pyspark.sql.utils import AnalysisException\n \n- try:\n- return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n- except AnalysisException:\n- # In serverless clusters, session.conf.get() is unsupported\n- # and raises an AnalysisException. We may encounter this case\n- # when DBConnect is used to connect to a serverless cluster,\n- # in which case the prior `is_in_databricks_serverless_runtime()`\n- # check will have returned false (as of 2025-06-07, it checks\n- # an environment variable that isn't set by DBConnect)\n- return None\n \n \n-def _get_registry_uri_from_context():\n- if _registry_uri is not None:\n- return _registry_uri\n- elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n- return uri\n- return _registry_uri\n \n \n-def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n- \"\"\"\n- Get the default registry URI for a given tracking URI.\n \n- If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n- Otherwise, returns the tracking URI itself.\n \n- Args:\n- tracking_uri: The tracking URI to get the default registry URI for\n \n- Returns:\n- The default registry URI\n- \"\"\"\n- if tracking_uri is not None and is_databricks_uri(tracking_uri):\n- # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n- # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n- # model registry offering on Databricks\n- if tracking_uri == \"databricks\":\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n- else:\n- # Extract profile from tracking URI and construct databricks-uc URI\n- profile, key_prefix = get_db_info_from_uri(tracking_uri)\n- if profile:\n- # Reconstruct the profile string including key_prefix if present\n- profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n- return construct_db_uc_uri_from_profile(profile_string)\n- else:\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n-\n- # For non-databricks tracking URIs, use the tracking URI as the registry URI\n- return tracking_uri\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n-def _resolve_registry_uri(\n- registry_uri: str | None = None, tracking_uri: str | None = None\n-) -> str | None:\n- \"\"\"\n- Resolve the registry URI following the same logic as get_registry_uri().\n- \"\"\"\n- return (\n- registry_uri\n- or _get_registry_uri_from_context()\n- or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n-def _get_rest_store(store_uri, **_):\n- return RestStore(partial(get_default_host_creds, store_uri))\n+\n+\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n-def _get_store_registry():\n- global _model_registry_store_registry\n- from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n- from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n- if _model_registry_store_registry is not None:\n- return _model_registry_store_registry\n \n- _model_registry_store_registry = ModelRegistryStoreRegistry()\n- _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n- # Register a placeholder function that raises if users pass a registry URI with scheme\n- # \"databricks-uc\"\n- _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n- _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n- for scheme in [\"http\", \"https\"]:\n- _model_registry_store_registry.register(scheme, _get_rest_store)\n \n- for scheme in DATABASE_ENGINES:\n- _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n- for scheme in [\"\", \"file\"]:\n- _model_registry_store_registry.register(scheme, _get_file_store)\n \n- _model_registry_store_registry.register_entrypoints()\n- return _model_registry_store_registry\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n--- a/mlflow/tracking/_tracking_service/utils.py\n+++ b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n- @property\n- def tracking_uri(self):\n- return self._tracking_client.tracking_uri\n \n- def _get_registry_client(self):\n- \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n- Raises:\n- MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n- example, when the registry URI refers to an unsupported store type (e.g., the\n- FileStore).\n \n- Returns:\n- A ModelRegistryClient instance.\n- \"\"\"\n- # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n- # an instance variable on this `MlflowClient` instance. Because the instance variable\n- # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n- # function is used to safely fetch the variable (if it is defined) or a NoneType\n- # (if it is not defined)\n- registry_client_attr = \"_registry_client_lazy\"\n- registry_client = getattr(self, registry_client_attr, None)\n- if registry_client is None:\n- try:\n- registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n- # Define an instance variable on this `MlflowClient` instance to reference the\n- # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n- # that the variable name is consistent with the variable name specified in the\n- # preceding call to `getattr()`\n- setattr(self, registry_client_attr, registry_client)\n- except UnsupportedModelRegistryStoreURIException as exc:\n- raise MlflowException(\n- \"Model Registry features are not supported by the store with URI:\"\n- f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n- f\" {exc.supported_uri_schemes}.\",\n- FEATURE_DISABLED,\n- )\n- return registry_client\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEventStr | WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of events that trigger this webhook. Can be strings or\n- `WebhookEvent` objects.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self._get_registry_client().get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self._get_registry_client().list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEventStr | WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of events. Can be strings or `WebhookEvent` objects.\n- secret: New webhook secret.\n- status: New webhook status. Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().update_webhook(\n- webhook_id=webhook_id,\n- name=name,\n- description=description,\n- url=url,\n- events=events,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self._get_registry_client().delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n- return self._get_registry_client().test_webhook(webhook_id, event)\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,7 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,36 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n--- a/mlflow/utils/credentials.py\n+++ b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n-def get_default_host_creds(store_uri):\n- creds = read_mlflow_creds()\n- return MlflowHostCreds(\n- host=store_uri,\n- username=creds.username,\n- password=creds.password,\n- token=MLFLOW_TRACKING_TOKEN.get(),\n- aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n- auth=MLFLOW_TRACKING_AUTH.get(),\n- ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n- client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n- server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n-def parse_dict(js_dict, message):\n- \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n- _stringify_all_experiment_ids(js_dict)\n- ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n+\n+\n+\n+\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n--- a/mlflow/utils/request_utils.py\n+++ b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n-@lru_cache(maxsize=64)\n-def _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- # To create a new Session object for each process, we use the process id as the cache key.\n- # This is to avoid sharing the same Session object across processes, which can lead to issues\n- # such as https://stackoverflow.com/q/3724900.\n- _pid,\n- respect_retry_after_header=True,\n-):\n- \"\"\"\n- This function should not be called directly. Instead, use `_get_request_session` below.\n- \"\"\"\n \n- retry_kwargs = {\n- \"total\": max_retries,\n- \"connect\": max_retries,\n- \"read\": max_retries,\n- \"redirect\": max_retries,\n- \"status\": max_retries,\n- \"status_forcelist\": retry_codes,\n- \"backoff_factor\": backoff_factor,\n- \"backoff_jitter\": backoff_jitter,\n- \"raise_on_status\": raise_on_status,\n- \"respect_retry_after_header\": respect_retry_after_header,\n- }\n- urllib3_version = Version(urllib3.__version__)\n- if urllib3_version >= Version(\"1.26.0\"):\n- retry_kwargs[\"allowed_methods\"] = None\n- else:\n- retry_kwargs[\"method_whitelist\"] = None\n-\n- if urllib3_version < Version(\"2.0\"):\n- retry = JitteredRetry(**retry_kwargs)\n- else:\n- retry = Retry(**retry_kwargs)\n- from mlflow.environment_variables import (\n- MLFLOW_HTTP_POOL_CONNECTIONS,\n- MLFLOW_HTTP_POOL_MAXSIZE,\n- )\n-\n- adapter = HTTPAdapter(\n- pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n- pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n- max_retries=retry,\n- )\n- session = requests.Session()\n- session.mount(\"https://\", adapter)\n- session.mount(\"http://\", adapter)\n- return session\n-\n-\n-def _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n-):\n- \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n- Args:\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n- Returns:\n- requests.Session object.\n \n- \"\"\"\n- return _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- _pid=os.getpid(),\n- respect_retry_after_header=respect_retry_after_header,\n- )\n \n \n-def _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status=True,\n- allow_redirects=None,\n- respect_retry_after_header=True,\n- **kwargs,\n-):\n- \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n- Args:\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- url: The target URL address for the HTTP request.\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- session = _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n- )\n \n- # the environment variable is hardcoded here to avoid importing mlflow.\n- # however, documentation is available in environment_variables.py\n- env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n- allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n- return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n-from mlflow.utils.proto_json_utils import parse_dict\n-from mlflow.utils.request_utils import (\n- _TRANSIENT_FAILURE_RESPONSE_CODES,\n- _get_http_response_with_retries,\n- augmented_raise_for_status, # noqa: F401\n- cloud_storage_http_request, # noqa: F401\n-)\n-from mlflow.utils.string_utils import strip_suffix\n+\n+from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n+\n+\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n \n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n \n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n-\n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n \n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n-\n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n-\n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n-\n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n-\n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n-\n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n-\n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n-def _can_parse_as_json_object(string):\n- try:\n- return isinstance(json.loads(string), dict)\n- except Exception:\n- return False\n+\n+\n+\n+\n+\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n-def verify_rest_response(response, endpoint):\n- \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n- # Handle Armeria-specific response case where response text is \"200 OK\"\n- # v1/traces endpoint might return empty response\n- if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n- response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n- return response\n-\n- # Handle non-200 status codes\n- if response.status_code != 200:\n- if _can_parse_as_json_object(response.text):\n- raise RestException(json.loads(response.text))\n- else:\n- base_msg = (\n- f\"API request to endpoint {endpoint} \"\n- f\"failed with error code {response.status_code} != 200\"\n- )\n- raise MlflowException(\n- f\"{base_msg}. Response body: '{response.text}'\",\n- error_code=get_error_code(response.status_code),\n- )\n \n- # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n- # response\n- if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n- base_msg = (\n- \"API request to endpoint was successful but the response body was not \"\n- \"in a valid JSON format\"\n- )\n- raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n-\n- return response\n-\n-\n-def _validate_max_retries(max_retries):\n- max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n-\n- if max_retry_limit < 0:\n- raise MlflowException(\n- message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n- \"Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if max_retries >= max_retry_limit:\n- raise MlflowException(\n- message=f\"The configured max_retries value ({max_retries}) is \"\n- f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if max_retries < 0:\n- raise MlflowException(\n- message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n-\n-def _validate_backoff_factor(backoff_factor):\n- max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n-\n- if max_backoff_factor_limit < 0:\n- raise MlflowException(\n- message=\"The current maximum backoff factor limit is invalid \"\n- f\"({max_backoff_factor_limit}). Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor >= max_backoff_factor_limit:\n- raise MlflowException(\n- message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n- \"of the maximum allowable backoff_factor limit \"\n- f\"({max_backoff_factor_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor < 0:\n- raise MlflowException(\n- message=\"The backoff_factor value must be either 0 a positive integer. \"\n- f\"Got {backoff_factor}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n-def call_endpoint(\n- host_creds,\n- endpoint,\n- method,\n- json_body,\n- response_proto,\n- extra_headers=None,\n- retry_timeout_seconds=None,\n-):\n- # Convert json string to json dictionary, to pass to requests\n- if json_body is not None:\n- json_body = json.loads(json_body)\n- call_kwargs = {\n- \"host_creds\": host_creds,\n- \"endpoint\": endpoint,\n- \"method\": method,\n- }\n- if extra_headers is not None:\n- call_kwargs[\"extra_headers\"] = extra_headers\n- if retry_timeout_seconds is not None:\n- call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n- if method == \"GET\":\n- call_kwargs[\"params\"] = json_body\n- response = http_request(**call_kwargs)\n- else:\n- call_kwargs[\"json\"] = json_body\n- response = http_request(**call_kwargs)\n \n- response = verify_rest_response(response, endpoint)\n- response_to_parse = response.text\n- try:\n- js_dict = json.loads(response_to_parse)\n- except json.JSONDecodeError:\n- if len(response_to_parse) > 50:\n- response_to_parse = response_to_parse[:50] + \"...\"\n- _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n- raise\n \n- parse_dict(js_dict=js_dict, message=response_proto)\n- return response_proto\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n- def __init__(\n- self,\n- host,\n- username=None,\n- password=None,\n- token=None,\n- aws_sigv4=False,\n- auth=None,\n- ignore_tls_verification=False,\n- client_cert_path=None,\n- server_cert_path=None,\n- use_databricks_sdk=False,\n- databricks_auth_profile=None,\n- client_id=None,\n- client_secret=None,\n- use_secret_scope_token=False,\n- ):\n- if not host:\n- raise MlflowException(\n- message=\"host is a required parameter for MlflowHostCreds\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if ignore_tls_verification and (server_cert_path is not None):\n- raise MlflowException(\n- message=(\n- \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n- \"must not be set! This error may have occurred because the \"\n- \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n- \"environment variables are both set - only one of these environment \"\n- \"variables may be set.\"\n- ),\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- self.host = host\n- self.username = username\n- self.password = password\n- self.token = token\n- self.aws_sigv4 = aws_sigv4\n- self.auth = auth\n- self.ignore_tls_verification = ignore_tls_verification\n- self.client_cert_path = client_cert_path\n- self.server_cert_path = server_cert_path\n- self.use_databricks_sdk = use_databricks_sdk\n- self.databricks_auth_profile = databricks_auth_profile\n- self.client_id = client_id\n- self.client_secret = client_secret\n- self.use_secret_scope_token = use_secret_scope_token\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,12 +712,11 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n- @property\n- def verify(self):\n- if self.use_databricks_sdk:\n- # Let databricks-sdk set HTTP request `verify` param.\n- return None\n- if self.server_cert_path is None:\n- return not self.ignore_tls_verification\n- else:\n- return self.server_cert_path\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n+++ a/mlflow/entities/webhook.py\n--- b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n+ def __eq__(self, other: object) -> bool:\n+ if not isinstance(other, WebhookEvent):\n+ return False\n+ return self.entity == other.entity and self.action == other.action\n-\n-\n-\n-\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n+ @property\n+ def secret(self) -> str | None:\n+ return self._secret\n-\n-\n-\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n+++ a/mlflow/server/handlers.py\n--- b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n+++ a/mlflow/store/artifact/azure_blob_artifact_repo.py\n--- b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+from mlflow.utils.credentials import get_default_host_creds\n-\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ augmented_raise_for_status,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(self, store_uri=None, tracking_uri=None):\n+ \"\"\"\n+ Empty constructor. This is deliberately not marked as abstract, else every derived class\n+ would be forced to create one.\n \n+ Args:\n+ store_uri: The model registry store URI.\n+ tracking_uri: URI of the current MLflow tracking server, used to perform operations\n+ like fetching source run metadata or downloading source run artifacts\n+ to support subsequently uploading them to the model registry storage\n+ location.\n+ \"\"\"\n+ # Create a thread lock to ensure thread safety when linking prompts to other entities,\n+ # since the default linking implementation reads and appends entity tags, which\n+ # is prone to concurrent modification issues\n+ self._prompt_link_lock = threading.RLock()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n+++ a/mlflow/store/model_registry/base_rest_store.py\n--- b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n+from mlflow.utils.rest_utils import (\n+ call_endpoint,\n+ call_endpoints,\n+)\n-from mlflow.utils.rest_utils import call_endpoints\n-\n-\n-\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n+++ a/mlflow/tracking/_model_registry/client.py\n--- b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ @record_usage_event(CreateWebhookEvent)\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Unique name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of event types that trigger this webhook.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE).\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ return self.store.create_webhook(name, url, events, description, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self.store.get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self.store.list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of event types.\n+ secret: New webhook secret.\n+ status: New webhook status.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self.store.delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n+ return self.store.test_webhook(webhook_id, event)\n- return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n+++ a/mlflow/tracking/_model_registry/registry.py\n--- b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n+ from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n-\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n+++ a/mlflow/tracking/_model_registry/utils.py\n--- b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n+def _get_registry_uri_from_spark_session():\n+ session = _get_active_spark_session()\n+ if session is None:\n+ return None\n \n+ if is_in_databricks_serverless_runtime():\n+ # Connected to Serverless\n+ return \"databricks-uc\"\n \n+ from pyspark.sql.utils import AnalysisException\n \n+ try:\n+ return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n+ except AnalysisException:\n+ # In serverless clusters, session.conf.get() is unsupported\n+ # and raises an AnalysisException. We may encounter this case\n+ # when DBConnect is used to connect to a serverless cluster,\n+ # in which case the prior `is_in_databricks_serverless_runtime()`\n+ # check will have returned false (as of 2025-06-07, it checks\n+ # an environment variable that isn't set by DBConnect)\n+ return None\n \n \n+def _get_registry_uri_from_context():\n+ if _registry_uri is not None:\n+ return _registry_uri\n+ elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n+ return uri\n+ return _registry_uri\n \n \n+def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n+ \"\"\"\n+ Get the default registry URI for a given tracking URI.\n \n+ If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n+ Otherwise, returns the tracking URI itself.\n \n+ Args:\n+ tracking_uri: The tracking URI to get the default registry URI for\n \n+ Returns:\n+ The default registry URI\n+ \"\"\"\n+ if tracking_uri is not None and is_databricks_uri(tracking_uri):\n+ # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n+ # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n+ # model registry offering on Databricks\n+ if tracking_uri == \"databricks\":\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+ else:\n+ # Extract profile from tracking URI and construct databricks-uc URI\n+ profile, key_prefix = get_db_info_from_uri(tracking_uri)\n+ if profile:\n+ # Reconstruct the profile string including key_prefix if present\n+ profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n+ return construct_db_uc_uri_from_profile(profile_string)\n+ else:\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+\n+ # For non-databricks tracking URIs, use the tracking URI as the registry URI\n+ return tracking_uri\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n+def _resolve_registry_uri(\n+ registry_uri: str | None = None, tracking_uri: str | None = None\n+) -> str | None:\n+ \"\"\"\n+ Resolve the registry URI following the same logic as get_registry_uri().\n+ \"\"\"\n+ return (\n+ registry_uri\n+ or _get_registry_uri_from_context()\n+ or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n+def _get_rest_store(store_uri, **_):\n+ return RestStore(partial(get_default_host_creds, store_uri))\n-\n-\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n+def _get_store_registry():\n+ global _model_registry_store_registry\n+ from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n+ from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n+ if _model_registry_store_registry is not None:\n+ return _model_registry_store_registry\n \n+ _model_registry_store_registry = ModelRegistryStoreRegistry()\n+ _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n+ # Register a placeholder function that raises if users pass a registry URI with scheme\n+ # \"databricks-uc\"\n+ _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n+ _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n+ for scheme in [\"http\", \"https\"]:\n+ _model_registry_store_registry.register(scheme, _get_rest_store)\n \n+ for scheme in DATABASE_ENGINES:\n+ _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n+ for scheme in [\"\", \"file\"]:\n+ _model_registry_store_registry.register(scheme, _get_file_store)\n \n+ _model_registry_store_registry.register_entrypoints()\n+ return _model_registry_store_registry\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n+++ a/mlflow/tracking/_tracking_service/utils.py\n--- b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n+++ a/mlflow/tracking/client.py\n--- b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n+ @property\n+ def tracking_uri(self):\n+ return self._tracking_client.tracking_uri\n \n+ def _get_registry_client(self):\n+ \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n+ Raises:\n+ MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n+ example, when the registry URI refers to an unsupported store type (e.g., the\n+ FileStore).\n \n+ Returns:\n+ A ModelRegistryClient instance.\n+ \"\"\"\n+ # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n+ # an instance variable on this `MlflowClient` instance. Because the instance variable\n+ # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n+ # function is used to safely fetch the variable (if it is defined) or a NoneType\n+ # (if it is not defined)\n+ registry_client_attr = \"_registry_client_lazy\"\n+ registry_client = getattr(self, registry_client_attr, None)\n+ if registry_client is None:\n+ try:\n+ registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n+ # Define an instance variable on this `MlflowClient` instance to reference the\n+ # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n+ # that the variable name is consistent with the variable name specified in the\n+ # preceding call to `getattr()`\n+ setattr(self, registry_client_attr, registry_client)\n+ except UnsupportedModelRegistryStoreURIException as exc:\n+ raise MlflowException(\n+ \"Model Registry features are not supported by the store with URI:\"\n+ f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n+ f\" {exc.supported_uri_schemes}.\",\n+ FEATURE_DISABLED,\n+ )\n+ return registry_client\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEventStr | WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of events that trigger this webhook. Can be strings or\n+ `WebhookEvent` objects.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self._get_registry_client().get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self._get_registry_client().list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEventStr | WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of events. Can be strings or `WebhookEvent` objects.\n+ secret: New webhook secret.\n+ status: New webhook status. Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ if events is not None:\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().update_webhook(\n+ webhook_id=webhook_id,\n+ name=name,\n+ description=description,\n+ url=url,\n+ events=events,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self._get_registry_client().delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n- return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,7 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,36 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n+++ a/mlflow/utils/credentials.py\n--- b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n+def get_default_host_creds(store_uri):\n+ creds = read_mlflow_creds()\n+ return MlflowHostCreds(\n+ host=store_uri,\n+ username=creds.username,\n+ password=creds.password,\n+ token=MLFLOW_TRACKING_TOKEN.get(),\n+ aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n+ auth=MLFLOW_TRACKING_AUTH.get(),\n+ ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n+ client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n+ server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n+def parse_dict(js_dict, message):\n+ \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n+ _stringify_all_experiment_ids(js_dict)\n+ ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n-\n-\n-\n-\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n+++ a/mlflow/utils/request_utils.py\n--- b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n+@lru_cache(maxsize=64)\n+def _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ # To create a new Session object for each process, we use the process id as the cache key.\n+ # This is to avoid sharing the same Session object across processes, which can lead to issues\n+ # such as https://stackoverflow.com/q/3724900.\n+ _pid,\n+ respect_retry_after_header=True,\n+):\n+ \"\"\"\n+ This function should not be called directly. Instead, use `_get_request_session` below.\n+ \"\"\"\n \n+ retry_kwargs = {\n+ \"total\": max_retries,\n+ \"connect\": max_retries,\n+ \"read\": max_retries,\n+ \"redirect\": max_retries,\n+ \"status\": max_retries,\n+ \"status_forcelist\": retry_codes,\n+ \"backoff_factor\": backoff_factor,\n+ \"backoff_jitter\": backoff_jitter,\n+ \"raise_on_status\": raise_on_status,\n+ \"respect_retry_after_header\": respect_retry_after_header,\n+ }\n+ urllib3_version = Version(urllib3.__version__)\n+ if urllib3_version >= Version(\"1.26.0\"):\n+ retry_kwargs[\"allowed_methods\"] = None\n+ else:\n+ retry_kwargs[\"method_whitelist\"] = None\n+\n+ if urllib3_version < Version(\"2.0\"):\n+ retry = JitteredRetry(**retry_kwargs)\n+ else:\n+ retry = Retry(**retry_kwargs)\n+ from mlflow.environment_variables import (\n+ MLFLOW_HTTP_POOL_CONNECTIONS,\n+ MLFLOW_HTTP_POOL_MAXSIZE,\n+ )\n+\n+ adapter = HTTPAdapter(\n+ pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n+ pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n+ max_retries=retry,\n+ )\n+ session = requests.Session()\n+ session.mount(\"https://\", adapter)\n+ session.mount(\"http://\", adapter)\n+ return session\n+\n+\n+def _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+):\n+ \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n+ Args:\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n+ Returns:\n+ requests.Session object.\n \n+ \"\"\"\n+ return _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ _pid=os.getpid(),\n+ respect_retry_after_header=respect_retry_after_header,\n+ )\n \n \n+def _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status=True,\n+ allow_redirects=None,\n+ respect_retry_after_header=True,\n+ **kwargs,\n+):\n+ \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n+ Args:\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ url: The target URL address for the HTTP request.\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ session = _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+ )\n \n+ # the environment variable is hardcoded here to avoid importing mlflow.\n+ # however, documentation is available in environment_variables.py\n+ env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n+ allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n+ return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n+from mlflow.utils.proto_json_utils import parse_dict\n+from mlflow.utils.request_utils import (\n+ _TRANSIENT_FAILURE_RESPONSE_CODES,\n+ _get_http_response_with_retries,\n+ augmented_raise_for_status, # noqa: F401\n+ cloud_storage_http_request, # noqa: F401\n+)\n+from mlflow.utils.string_utils import strip_suffix\n-\n-from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n-\n-\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n \n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n \n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n+\n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n \n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n+\n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n+\n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n+\n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n+\n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n+\n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n+\n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n+def _can_parse_as_json_object(string):\n+ try:\n+ return isinstance(json.loads(string), dict)\n+ except Exception:\n+ return False\n-\n-\n-\n-\n-\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n+def verify_rest_response(response, endpoint):\n+ \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n+ # Handle Armeria-specific response case where response text is \"200 OK\"\n+ # v1/traces endpoint might return empty response\n+ if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n+ response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n+ return response\n+\n+ # Handle non-200 status codes\n+ if response.status_code != 200:\n+ if _can_parse_as_json_object(response.text):\n+ raise RestException(json.loads(response.text))\n+ else:\n+ base_msg = (\n+ f\"API request to endpoint {endpoint} \"\n+ f\"failed with error code {response.status_code} != 200\"\n+ )\n+ raise MlflowException(\n+ f\"{base_msg}. Response body: '{response.text}'\",\n+ error_code=get_error_code(response.status_code),\n+ )\n \n+ # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n+ # response\n+ if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n+ base_msg = (\n+ \"API request to endpoint was successful but the response body was not \"\n+ \"in a valid JSON format\"\n+ )\n+ raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n+\n+ return response\n+\n+\n+def _validate_max_retries(max_retries):\n+ max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n+\n+ if max_retry_limit < 0:\n+ raise MlflowException(\n+ message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n+ \"Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if max_retries >= max_retry_limit:\n+ raise MlflowException(\n+ message=f\"The configured max_retries value ({max_retries}) is \"\n+ f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if max_retries < 0:\n+ raise MlflowException(\n+ message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+\n+def _validate_backoff_factor(backoff_factor):\n+ max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n+\n+ if max_backoff_factor_limit < 0:\n+ raise MlflowException(\n+ message=\"The current maximum backoff factor limit is invalid \"\n+ f\"({max_backoff_factor_limit}). Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor >= max_backoff_factor_limit:\n+ raise MlflowException(\n+ message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n+ \"of the maximum allowable backoff_factor limit \"\n+ f\"({max_backoff_factor_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor < 0:\n+ raise MlflowException(\n+ message=\"The backoff_factor value must be either 0 a positive integer. \"\n+ f\"Got {backoff_factor}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n+def call_endpoint(\n+ host_creds,\n+ endpoint,\n+ method,\n+ json_body,\n+ response_proto,\n+ extra_headers=None,\n+ retry_timeout_seconds=None,\n+):\n+ # Convert json string to json dictionary, to pass to requests\n+ if json_body is not None:\n+ json_body = json.loads(json_body)\n+ call_kwargs = {\n+ \"host_creds\": host_creds,\n+ \"endpoint\": endpoint,\n+ \"method\": method,\n+ }\n+ if extra_headers is not None:\n+ call_kwargs[\"extra_headers\"] = extra_headers\n+ if retry_timeout_seconds is not None:\n+ call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n+ if method == \"GET\":\n+ call_kwargs[\"params\"] = json_body\n+ response = http_request(**call_kwargs)\n+ else:\n+ call_kwargs[\"json\"] = json_body\n+ response = http_request(**call_kwargs)\n \n+ response = verify_rest_response(response, endpoint)\n+ response_to_parse = response.text\n+ try:\n+ js_dict = json.loads(response_to_parse)\n+ except json.JSONDecodeError:\n+ if len(response_to_parse) > 50:\n+ response_to_parse = response_to_parse[:50] + \"...\"\n+ _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n+ raise\n \n+ parse_dict(js_dict=js_dict, message=response_proto)\n+ return response_proto\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n+ def __init__(\n+ self,\n+ host,\n+ username=None,\n+ password=None,\n+ token=None,\n+ aws_sigv4=False,\n+ auth=None,\n+ ignore_tls_verification=False,\n+ client_cert_path=None,\n+ server_cert_path=None,\n+ use_databricks_sdk=False,\n+ databricks_auth_profile=None,\n+ client_id=None,\n+ client_secret=None,\n+ use_secret_scope_token=False,\n+ ):\n+ if not host:\n+ raise MlflowException(\n+ message=\"host is a required parameter for MlflowHostCreds\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if ignore_tls_verification and (server_cert_path is not None):\n+ raise MlflowException(\n+ message=(\n+ \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n+ \"must not be set! This error may have occurred because the \"\n+ \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n+ \"environment variables are both set - only one of these environment \"\n+ \"variables may be set.\"\n+ ),\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ self.host = host\n+ self.username = username\n+ self.password = password\n+ self.token = token\n+ self.aws_sigv4 = aws_sigv4\n+ self.auth = auth\n+ self.ignore_tls_verification = ignore_tls_verification\n+ self.client_cert_path = client_cert_path\n+ self.server_cert_path = server_cert_path\n+ self.use_databricks_sdk = use_databricks_sdk\n+ self.databricks_auth_profile = databricks_auth_profile\n+ self.client_id = client_id\n+ self.client_secret = client_secret\n+ self.use_secret_scope_token = use_secret_scope_token\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,11 +712,12 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n+ @property\n+ def verify(self):\n+ if self.use_databricks_sdk:\n+ # Let databricks-sdk set HTTP request `verify` param.\n+ return None\n+ if self.server_cert_path is None:\n+ return not self.ignore_tls_verification\n+ else:\n+ return self.server_cert_path\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..20dc8f033 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -6337,14 +6337,17 @@ class MlflowClient:\n if status is not None:\n status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n+ try:\n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n+ except MlflowException:\n+ return None\n \n @experimental(version=\"3.3.0\")\n def get_webhook(self, webhook_id: str) -> Webhook:\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py\ndeleted file mode 100644\nindex 97fd61b00..000000000\n--- a/tests/tracking/test_client_webhooks.py\n+++ /dev/null\n@@ -1,219 +0,0 @@\n-from pathlib import Path\n-from typing import Iterator\n-\n-import pytest\n-from cryptography.fernet import Fernet\n-\n-from mlflow.entities.webhook import WebhookAction, WebhookEntity, WebhookEvent, WebhookStatus\n-from mlflow.environment_variables import MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY\n-from mlflow.exceptions import MlflowException\n-from mlflow.server import handlers\n-from mlflow.server.fastapi_app import app\n-from mlflow.server.handlers import initialize_backend_stores\n-from mlflow.tracking import MlflowClient\n-\n-from tests.helper_functions import get_safe_port\n-from tests.tracking.integration_test_utils import ServerThread\n-\n-\n-@pytest.fixture\n-def client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Iterator[MlflowClient]:\n- \"\"\"Setup a local MLflow server with proper webhook encryption key support.\"\"\"\n- # Set up encryption key for webhooks using monkeypatch\n- encryption_key = Fernet.generate_key().decode(\"utf-8\")\n- monkeypatch.setenv(MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY.name, encryption_key)\n-\n- # Configure backend stores\n- backend_uri = f\"sqlite:///{tmp_path / 'mlflow.db'}\"\n- default_artifact_root = tmp_path.as_uri()\n-\n- # Force-reset backend stores before each test\n- handlers._tracking_store = None\n- handlers._model_registry_store = None\n- initialize_backend_stores(backend_uri, default_artifact_root=default_artifact_root)\n-\n- # Start server and return client\n- with ServerThread(app, get_safe_port()) as url:\n- yield MlflowClient(url)\n-\n-\n-def test_create_webhook(client: MlflowClient):\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n- assert webhook.events == [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n-\n- webhook = client.get_webhook(webhook.webhook_id)\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n-\n- # With secret\n- webhook_with_secret = client.create_webhook(\n- name=\"test_webhook_with_secret\",\n- url=\"https://example.com/webhook_with_secret\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- secret=\"my_secret\",\n- )\n- assert webhook_with_secret.name == \"test_webhook_with_secret\"\n- assert webhook_with_secret.url == \"https://example.com/webhook_with_secret\"\n- assert webhook_with_secret.secret is None\n- assert webhook_with_secret.events == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)\n- ]\n-\n- # Multiple events\n- webhook_multiple_events = client.create_webhook(\n- name=\"test_webhook_multiple_events\",\n- url=\"https://example.com/webhook_multiple_events\",\n- events=[\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- ],\n- )\n- assert webhook_multiple_events.name == \"test_webhook_multiple_events\"\n- assert webhook_multiple_events.url == \"https://example.com/webhook_multiple_events\"\n- assert sorted(\n- webhook_multiple_events.events, key=lambda e: (e.entity.value, e.action.value)\n- ) == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- ]\n- assert webhook_multiple_events.secret is None\n-\n-\n-def test_get_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- created_webhook = client.create_webhook(\n- name=\"test_webhook\", url=\"https://example.com/webhook\", events=events\n- )\n- retrieved_webhook = client.get_webhook(created_webhook.webhook_id)\n- assert retrieved_webhook.webhook_id == created_webhook.webhook_id\n- assert retrieved_webhook.name == \"test_webhook\"\n- assert retrieved_webhook.url == \"https://example.com/webhook\"\n- assert retrieved_webhook.events == events\n-\n-\n-def test_get_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.get_webhook(\"nonexistent\")\n-\n-\n-def test_list_webhooks(client: MlflowClient):\n- # Create more webhooks than max_results\n- created_webhooks = []\n- for i in range(5):\n- webhook = client.create_webhook(\n- name=f\"webhook{i}\",\n- url=f\"https://example.com/{i}\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- created_webhooks.append(webhook)\n- # Test pagination with max_results=2\n- webhooks_page = client.list_webhooks(max_results=2)\n- assert len(webhooks_page) == 2\n- assert webhooks_page.token is not None\n- # Get next page\n- next_webhooks_page = client.list_webhooks(max_results=2, page_token=webhooks_page.token)\n- assert len(next_webhooks_page) == 2\n- assert next_webhooks_page.token is not None\n- # Verify we don't get duplicates\n- first_page_ids = {w.webhook_id for w in webhooks_page}\n- second_page_ids = {w.webhook_id for w in next_webhooks_page}\n- assert first_page_ids.isdisjoint(second_page_ids)\n-\n-\n-def test_update_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\", url=\"https://example.com/original\", events=events\n- )\n- # Update webhook\n- new_events = [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.REGISTERED_MODEL, WebhookAction.CREATED),\n- ]\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- url=\"https://example.com/updated\",\n- events=new_events,\n- description=\"Updated description\",\n- secret=\"new_secret\",\n- status=WebhookStatus.DISABLED,\n- )\n- assert updated_webhook.webhook_id == webhook.webhook_id\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/updated\"\n- assert updated_webhook.events == new_events\n- assert updated_webhook.description == \"Updated description\"\n- assert updated_webhook.status == WebhookStatus.DISABLED\n- assert updated_webhook.last_updated_timestamp > webhook.last_updated_timestamp\n-\n-\n-def test_update_webhook_partial(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\",\n- url=\"https://example.com/original\",\n- events=events,\n- description=\"Original description\",\n- )\n- # Update only the name\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- )\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/original\"\n- assert updated_webhook.events == events\n- assert updated_webhook.description == \"Original description\"\n-\n-\n-def test_update_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.update_webhook(webhook_id=\"nonexistent\", name=\"new_name\")\n-\n-\n-@pytest.mark.parametrize(\n- (\"invalid_url\", \"expected_match\"),\n- [\n- (\" \", r\"Webhook URL cannot be empty or just whitespace\"),\n- (\"ftp://example.com\", r\"Invalid webhook URL scheme\"),\n- (\"http://[invalid\", r\"Invalid webhook URL\"),\n- ],\n-)\n-def test_update_webhook_invalid_urls(client: MlflowClient, invalid_url: str, expected_match: str):\n- # Create a valid webhook first\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- with pytest.raises(MlflowException, match=expected_match):\n- client.update_webhook(webhook_id=webhook.webhook_id, url=invalid_url)\n-\n-\n-def test_delete_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=events,\n- )\n- client.delete_webhook(webhook.webhook_id)\n- with pytest.raises(MlflowException, match=r\"Webhook with ID .* not found\"):\n- client.get_webhook(webhook.webhook_id)\n- webhooks_page = client.list_webhooks()\n- webhook_ids = {w.webhook_id for w in webhooks_page}\n- assert webhook.webhook_id not in webhook_ids\n-\n-\n-def test_delete_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.delete_webhook(\"nonexistent\")\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 2,\n \"shared_lines\": 5\n },\n {\n \"spec_id\": \"S3\",\n \"shared_functions\": 2,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S2\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":57,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S6\",\n \"text\": \"Raises: mlflow.MlflowException: If the webhook with the specified ID does not exist or if there are permission issues accessing the webhook.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracking/client.py\",\n \"lines\": \"6366\",\n \"function\": \"MlflowClient.get_webhook\"\n },\n {\n \"file\": \"mlflow/tracking/_model_registry/client.py\",\n \"lines\": \"836\",\n \"function\": \"ModelRegistryClient.get_webhook\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n--- a/mlflow/entities/webhook.py\n+++ b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n- def __eq__(self, other: object) -> bool:\n- if not isinstance(other, WebhookEvent):\n- return False\n- return self.entity == other.entity and self.action == other.action\n+\n+\n+\n+\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n- @property\n- def secret(self) -> str | None:\n- return self._secret\n+\n+\n+\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n--- a/mlflow/server/handlers.py\n+++ b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n--- a/mlflow/store/artifact/azure_blob_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-from mlflow.utils.credentials import get_default_host_creds\n+\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- augmented_raise_for_status,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n- def __init__(self, store_uri=None, tracking_uri=None):\n- \"\"\"\n- Empty constructor. This is deliberately not marked as abstract, else every derived class\n- would be forced to create one.\n \n- Args:\n- store_uri: The model registry store URI.\n- tracking_uri: URI of the current MLflow tracking server, used to perform operations\n- like fetching source run metadata or downloading source run artifacts\n- to support subsequently uploading them to the model registry storage\n- location.\n- \"\"\"\n- # Create a thread lock to ensure thread safety when linking prompts to other entities,\n- # since the default linking implementation reads and appends entity tags, which\n- # is prone to concurrent modification issues\n- self._prompt_link_lock = threading.RLock()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n--- a/mlflow/store/model_registry/base_rest_store.py\n+++ b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n-from mlflow.utils.rest_utils import (\n- call_endpoint,\n- call_endpoints,\n-)\n+from mlflow.utils.rest_utils import call_endpoints\n+\n+\n+\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- @record_usage_event(CreateWebhookEvent)\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Unique name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of event types that trigger this webhook.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE).\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- return self.store.create_webhook(name, url, events, description, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self.store.get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self.store.list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of event types.\n- secret: New webhook secret.\n- status: New webhook status.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self.store.delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n- return self.store.test_webhook(webhook_id, event)\n+ return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n--- a/mlflow/tracking/_model_registry/registry.py\n+++ b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n- from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n+\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n--- a/mlflow/tracking/_model_registry/utils.py\n+++ b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n-def _get_registry_uri_from_spark_session():\n- session = _get_active_spark_session()\n- if session is None:\n- return None\n \n- if is_in_databricks_serverless_runtime():\n- # Connected to Serverless\n- return \"databricks-uc\"\n \n- from pyspark.sql.utils import AnalysisException\n \n- try:\n- return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n- except AnalysisException:\n- # In serverless clusters, session.conf.get() is unsupported\n- # and raises an AnalysisException. We may encounter this case\n- # when DBConnect is used to connect to a serverless cluster,\n- # in which case the prior `is_in_databricks_serverless_runtime()`\n- # check will have returned false (as of 2025-06-07, it checks\n- # an environment variable that isn't set by DBConnect)\n- return None\n \n \n-def _get_registry_uri_from_context():\n- if _registry_uri is not None:\n- return _registry_uri\n- elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n- return uri\n- return _registry_uri\n \n \n-def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n- \"\"\"\n- Get the default registry URI for a given tracking URI.\n \n- If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n- Otherwise, returns the tracking URI itself.\n \n- Args:\n- tracking_uri: The tracking URI to get the default registry URI for\n \n- Returns:\n- The default registry URI\n- \"\"\"\n- if tracking_uri is not None and is_databricks_uri(tracking_uri):\n- # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n- # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n- # model registry offering on Databricks\n- if tracking_uri == \"databricks\":\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n- else:\n- # Extract profile from tracking URI and construct databricks-uc URI\n- profile, key_prefix = get_db_info_from_uri(tracking_uri)\n- if profile:\n- # Reconstruct the profile string including key_prefix if present\n- profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n- return construct_db_uc_uri_from_profile(profile_string)\n- else:\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n-\n- # For non-databricks tracking URIs, use the tracking URI as the registry URI\n- return tracking_uri\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n-def _resolve_registry_uri(\n- registry_uri: str | None = None, tracking_uri: str | None = None\n-) -> str | None:\n- \"\"\"\n- Resolve the registry URI following the same logic as get_registry_uri().\n- \"\"\"\n- return (\n- registry_uri\n- or _get_registry_uri_from_context()\n- or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n-def _get_rest_store(store_uri, **_):\n- return RestStore(partial(get_default_host_creds, store_uri))\n+\n+\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n-def _get_store_registry():\n- global _model_registry_store_registry\n- from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n- from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n- if _model_registry_store_registry is not None:\n- return _model_registry_store_registry\n \n- _model_registry_store_registry = ModelRegistryStoreRegistry()\n- _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n- # Register a placeholder function that raises if users pass a registry URI with scheme\n- # \"databricks-uc\"\n- _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n- _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n- for scheme in [\"http\", \"https\"]:\n- _model_registry_store_registry.register(scheme, _get_rest_store)\n \n- for scheme in DATABASE_ENGINES:\n- _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n- for scheme in [\"\", \"file\"]:\n- _model_registry_store_registry.register(scheme, _get_file_store)\n \n- _model_registry_store_registry.register_entrypoints()\n- return _model_registry_store_registry\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n--- a/mlflow/tracking/_tracking_service/utils.py\n+++ b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n- @property\n- def tracking_uri(self):\n- return self._tracking_client.tracking_uri\n \n- def _get_registry_client(self):\n- \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n- Raises:\n- MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n- example, when the registry URI refers to an unsupported store type (e.g., the\n- FileStore).\n \n- Returns:\n- A ModelRegistryClient instance.\n- \"\"\"\n- # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n- # an instance variable on this `MlflowClient` instance. Because the instance variable\n- # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n- # function is used to safely fetch the variable (if it is defined) or a NoneType\n- # (if it is not defined)\n- registry_client_attr = \"_registry_client_lazy\"\n- registry_client = getattr(self, registry_client_attr, None)\n- if registry_client is None:\n- try:\n- registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n- # Define an instance variable on this `MlflowClient` instance to reference the\n- # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n- # that the variable name is consistent with the variable name specified in the\n- # preceding call to `getattr()`\n- setattr(self, registry_client_attr, registry_client)\n- except UnsupportedModelRegistryStoreURIException as exc:\n- raise MlflowException(\n- \"Model Registry features are not supported by the store with URI:\"\n- f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n- f\" {exc.supported_uri_schemes}.\",\n- FEATURE_DISABLED,\n- )\n- return registry_client\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEventStr | WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of events that trigger this webhook. Can be strings or\n- `WebhookEvent` objects.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self._get_registry_client().get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self._get_registry_client().list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEventStr | WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of events. Can be strings or `WebhookEvent` objects.\n- secret: New webhook secret.\n- status: New webhook status. Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().update_webhook(\n- webhook_id=webhook_id,\n- name=name,\n- description=description,\n- url=url,\n- events=events,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self._get_registry_client().delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n- return self._get_registry_client().test_webhook(webhook_id, event)\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,7 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,36 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n--- a/mlflow/utils/credentials.py\n+++ b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n-def get_default_host_creds(store_uri):\n- creds = read_mlflow_creds()\n- return MlflowHostCreds(\n- host=store_uri,\n- username=creds.username,\n- password=creds.password,\n- token=MLFLOW_TRACKING_TOKEN.get(),\n- aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n- auth=MLFLOW_TRACKING_AUTH.get(),\n- ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n- client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n- server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n-def parse_dict(js_dict, message):\n- \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n- _stringify_all_experiment_ids(js_dict)\n- ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n+\n+\n+\n+\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n--- a/mlflow/utils/request_utils.py\n+++ b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n-@lru_cache(maxsize=64)\n-def _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- # To create a new Session object for each process, we use the process id as the cache key.\n- # This is to avoid sharing the same Session object across processes, which can lead to issues\n- # such as https://stackoverflow.com/q/3724900.\n- _pid,\n- respect_retry_after_header=True,\n-):\n- \"\"\"\n- This function should not be called directly. Instead, use `_get_request_session` below.\n- \"\"\"\n \n- retry_kwargs = {\n- \"total\": max_retries,\n- \"connect\": max_retries,\n- \"read\": max_retries,\n- \"redirect\": max_retries,\n- \"status\": max_retries,\n- \"status_forcelist\": retry_codes,\n- \"backoff_factor\": backoff_factor,\n- \"backoff_jitter\": backoff_jitter,\n- \"raise_on_status\": raise_on_status,\n- \"respect_retry_after_header\": respect_retry_after_header,\n- }\n- urllib3_version = Version(urllib3.__version__)\n- if urllib3_version >= Version(\"1.26.0\"):\n- retry_kwargs[\"allowed_methods\"] = None\n- else:\n- retry_kwargs[\"method_whitelist\"] = None\n-\n- if urllib3_version < Version(\"2.0\"):\n- retry = JitteredRetry(**retry_kwargs)\n- else:\n- retry = Retry(**retry_kwargs)\n- from mlflow.environment_variables import (\n- MLFLOW_HTTP_POOL_CONNECTIONS,\n- MLFLOW_HTTP_POOL_MAXSIZE,\n- )\n-\n- adapter = HTTPAdapter(\n- pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n- pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n- max_retries=retry,\n- )\n- session = requests.Session()\n- session.mount(\"https://\", adapter)\n- session.mount(\"http://\", adapter)\n- return session\n-\n-\n-def _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n-):\n- \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n- Args:\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n- Returns:\n- requests.Session object.\n \n- \"\"\"\n- return _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- _pid=os.getpid(),\n- respect_retry_after_header=respect_retry_after_header,\n- )\n \n \n-def _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status=True,\n- allow_redirects=None,\n- respect_retry_after_header=True,\n- **kwargs,\n-):\n- \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n- Args:\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- url: The target URL address for the HTTP request.\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- session = _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n- )\n \n- # the environment variable is hardcoded here to avoid importing mlflow.\n- # however, documentation is available in environment_variables.py\n- env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n- allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n- return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n-from mlflow.utils.proto_json_utils import parse_dict\n-from mlflow.utils.request_utils import (\n- _TRANSIENT_FAILURE_RESPONSE_CODES,\n- _get_http_response_with_retries,\n- augmented_raise_for_status, # noqa: F401\n- cloud_storage_http_request, # noqa: F401\n-)\n-from mlflow.utils.string_utils import strip_suffix\n+\n+from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n+\n+\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n \n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n \n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n-\n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n \n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n-\n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n-\n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n-\n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n-\n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n-\n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n-\n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n-def _can_parse_as_json_object(string):\n- try:\n- return isinstance(json.loads(string), dict)\n- except Exception:\n- return False\n+\n+\n+\n+\n+\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n-def verify_rest_response(response, endpoint):\n- \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n- # Handle Armeria-specific response case where response text is \"200 OK\"\n- # v1/traces endpoint might return empty response\n- if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n- response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n- return response\n-\n- # Handle non-200 status codes\n- if response.status_code != 200:\n- if _can_parse_as_json_object(response.text):\n- raise RestException(json.loads(response.text))\n- else:\n- base_msg = (\n- f\"API request to endpoint {endpoint} \"\n- f\"failed with error code {response.status_code} != 200\"\n- )\n- raise MlflowException(\n- f\"{base_msg}. Response body: '{response.text}'\",\n- error_code=get_error_code(response.status_code),\n- )\n \n- # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n- # response\n- if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n- base_msg = (\n- \"API request to endpoint was successful but the response body was not \"\n- \"in a valid JSON format\"\n- )\n- raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n-\n- return response\n-\n-\n-def _validate_max_retries(max_retries):\n- max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n-\n- if max_retry_limit < 0:\n- raise MlflowException(\n- message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n- \"Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if max_retries >= max_retry_limit:\n- raise MlflowException(\n- message=f\"The configured max_retries value ({max_retries}) is \"\n- f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if max_retries < 0:\n- raise MlflowException(\n- message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n-\n-def _validate_backoff_factor(backoff_factor):\n- max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n-\n- if max_backoff_factor_limit < 0:\n- raise MlflowException(\n- message=\"The current maximum backoff factor limit is invalid \"\n- f\"({max_backoff_factor_limit}). Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor >= max_backoff_factor_limit:\n- raise MlflowException(\n- message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n- \"of the maximum allowable backoff_factor limit \"\n- f\"({max_backoff_factor_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor < 0:\n- raise MlflowException(\n- message=\"The backoff_factor value must be either 0 a positive integer. \"\n- f\"Got {backoff_factor}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n-def call_endpoint(\n- host_creds,\n- endpoint,\n- method,\n- json_body,\n- response_proto,\n- extra_headers=None,\n- retry_timeout_seconds=None,\n-):\n- # Convert json string to json dictionary, to pass to requests\n- if json_body is not None:\n- json_body = json.loads(json_body)\n- call_kwargs = {\n- \"host_creds\": host_creds,\n- \"endpoint\": endpoint,\n- \"method\": method,\n- }\n- if extra_headers is not None:\n- call_kwargs[\"extra_headers\"] = extra_headers\n- if retry_timeout_seconds is not None:\n- call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n- if method == \"GET\":\n- call_kwargs[\"params\"] = json_body\n- response = http_request(**call_kwargs)\n- else:\n- call_kwargs[\"json\"] = json_body\n- response = http_request(**call_kwargs)\n \n- response = verify_rest_response(response, endpoint)\n- response_to_parse = response.text\n- try:\n- js_dict = json.loads(response_to_parse)\n- except json.JSONDecodeError:\n- if len(response_to_parse) > 50:\n- response_to_parse = response_to_parse[:50] + \"...\"\n- _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n- raise\n \n- parse_dict(js_dict=js_dict, message=response_proto)\n- return response_proto\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n- def __init__(\n- self,\n- host,\n- username=None,\n- password=None,\n- token=None,\n- aws_sigv4=False,\n- auth=None,\n- ignore_tls_verification=False,\n- client_cert_path=None,\n- server_cert_path=None,\n- use_databricks_sdk=False,\n- databricks_auth_profile=None,\n- client_id=None,\n- client_secret=None,\n- use_secret_scope_token=False,\n- ):\n- if not host:\n- raise MlflowException(\n- message=\"host is a required parameter for MlflowHostCreds\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if ignore_tls_verification and (server_cert_path is not None):\n- raise MlflowException(\n- message=(\n- \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n- \"must not be set! This error may have occurred because the \"\n- \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n- \"environment variables are both set - only one of these environment \"\n- \"variables may be set.\"\n- ),\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- self.host = host\n- self.username = username\n- self.password = password\n- self.token = token\n- self.aws_sigv4 = aws_sigv4\n- self.auth = auth\n- self.ignore_tls_verification = ignore_tls_verification\n- self.client_cert_path = client_cert_path\n- self.server_cert_path = server_cert_path\n- self.use_databricks_sdk = use_databricks_sdk\n- self.databricks_auth_profile = databricks_auth_profile\n- self.client_id = client_id\n- self.client_secret = client_secret\n- self.use_secret_scope_token = use_secret_scope_token\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,12 +712,11 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n- @property\n- def verify(self):\n- if self.use_databricks_sdk:\n- # Let databricks-sdk set HTTP request `verify` param.\n- return None\n- if self.server_cert_path is None:\n- return not self.ignore_tls_verification\n- else:\n- return self.server_cert_path\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n+++ a/mlflow/entities/webhook.py\n--- b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n+ def __eq__(self, other: object) -> bool:\n+ if not isinstance(other, WebhookEvent):\n+ return False\n+ return self.entity == other.entity and self.action == other.action\n-\n-\n-\n-\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n+ @property\n+ def secret(self) -> str | None:\n+ return self._secret\n-\n-\n-\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n+++ a/mlflow/server/handlers.py\n--- b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n+++ a/mlflow/store/artifact/azure_blob_artifact_repo.py\n--- b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+from mlflow.utils.credentials import get_default_host_creds\n-\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ augmented_raise_for_status,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(self, store_uri=None, tracking_uri=None):\n+ \"\"\"\n+ Empty constructor. This is deliberately not marked as abstract, else every derived class\n+ would be forced to create one.\n \n+ Args:\n+ store_uri: The model registry store URI.\n+ tracking_uri: URI of the current MLflow tracking server, used to perform operations\n+ like fetching source run metadata or downloading source run artifacts\n+ to support subsequently uploading them to the model registry storage\n+ location.\n+ \"\"\"\n+ # Create a thread lock to ensure thread safety when linking prompts to other entities,\n+ # since the default linking implementation reads and appends entity tags, which\n+ # is prone to concurrent modification issues\n+ self._prompt_link_lock = threading.RLock()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n+++ a/mlflow/store/model_registry/base_rest_store.py\n--- b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n+from mlflow.utils.rest_utils import (\n+ call_endpoint,\n+ call_endpoints,\n+)\n-from mlflow.utils.rest_utils import call_endpoints\n-\n-\n-\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n+++ a/mlflow/tracking/_model_registry/client.py\n--- b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ @record_usage_event(CreateWebhookEvent)\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Unique name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of event types that trigger this webhook.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE).\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ return self.store.create_webhook(name, url, events, description, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self.store.get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self.store.list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of event types.\n+ secret: New webhook secret.\n+ status: New webhook status.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self.store.delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n+ return self.store.test_webhook(webhook_id, event)\n- return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n+++ a/mlflow/tracking/_model_registry/registry.py\n--- b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n+ from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n-\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n+++ a/mlflow/tracking/_model_registry/utils.py\n--- b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n+def _get_registry_uri_from_spark_session():\n+ session = _get_active_spark_session()\n+ if session is None:\n+ return None\n \n+ if is_in_databricks_serverless_runtime():\n+ # Connected to Serverless\n+ return \"databricks-uc\"\n \n+ from pyspark.sql.utils import AnalysisException\n \n+ try:\n+ return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n+ except AnalysisException:\n+ # In serverless clusters, session.conf.get() is unsupported\n+ # and raises an AnalysisException. We may encounter this case\n+ # when DBConnect is used to connect to a serverless cluster,\n+ # in which case the prior `is_in_databricks_serverless_runtime()`\n+ # check will have returned false (as of 2025-06-07, it checks\n+ # an environment variable that isn't set by DBConnect)\n+ return None\n \n \n+def _get_registry_uri_from_context():\n+ if _registry_uri is not None:\n+ return _registry_uri\n+ elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n+ return uri\n+ return _registry_uri\n \n \n+def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n+ \"\"\"\n+ Get the default registry URI for a given tracking URI.\n \n+ If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n+ Otherwise, returns the tracking URI itself.\n \n+ Args:\n+ tracking_uri: The tracking URI to get the default registry URI for\n \n+ Returns:\n+ The default registry URI\n+ \"\"\"\n+ if tracking_uri is not None and is_databricks_uri(tracking_uri):\n+ # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n+ # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n+ # model registry offering on Databricks\n+ if tracking_uri == \"databricks\":\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+ else:\n+ # Extract profile from tracking URI and construct databricks-uc URI\n+ profile, key_prefix = get_db_info_from_uri(tracking_uri)\n+ if profile:\n+ # Reconstruct the profile string including key_prefix if present\n+ profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n+ return construct_db_uc_uri_from_profile(profile_string)\n+ else:\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+\n+ # For non-databricks tracking URIs, use the tracking URI as the registry URI\n+ return tracking_uri\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n+def _resolve_registry_uri(\n+ registry_uri: str | None = None, tracking_uri: str | None = None\n+) -> str | None:\n+ \"\"\"\n+ Resolve the registry URI following the same logic as get_registry_uri().\n+ \"\"\"\n+ return (\n+ registry_uri\n+ or _get_registry_uri_from_context()\n+ or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n+def _get_rest_store(store_uri, **_):\n+ return RestStore(partial(get_default_host_creds, store_uri))\n-\n-\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n+def _get_store_registry():\n+ global _model_registry_store_registry\n+ from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n+ from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n+ if _model_registry_store_registry is not None:\n+ return _model_registry_store_registry\n \n+ _model_registry_store_registry = ModelRegistryStoreRegistry()\n+ _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n+ # Register a placeholder function that raises if users pass a registry URI with scheme\n+ # \"databricks-uc\"\n+ _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n+ _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n+ for scheme in [\"http\", \"https\"]:\n+ _model_registry_store_registry.register(scheme, _get_rest_store)\n \n+ for scheme in DATABASE_ENGINES:\n+ _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n+ for scheme in [\"\", \"file\"]:\n+ _model_registry_store_registry.register(scheme, _get_file_store)\n \n+ _model_registry_store_registry.register_entrypoints()\n+ return _model_registry_store_registry\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n+++ a/mlflow/tracking/_tracking_service/utils.py\n--- b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n+++ a/mlflow/tracking/client.py\n--- b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n+ @property\n+ def tracking_uri(self):\n+ return self._tracking_client.tracking_uri\n \n+ def _get_registry_client(self):\n+ \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n+ Raises:\n+ MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n+ example, when the registry URI refers to an unsupported store type (e.g., the\n+ FileStore).\n \n+ Returns:\n+ A ModelRegistryClient instance.\n+ \"\"\"\n+ # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n+ # an instance variable on this `MlflowClient` instance. Because the instance variable\n+ # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n+ # function is used to safely fetch the variable (if it is defined) or a NoneType\n+ # (if it is not defined)\n+ registry_client_attr = \"_registry_client_lazy\"\n+ registry_client = getattr(self, registry_client_attr, None)\n+ if registry_client is None:\n+ try:\n+ registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n+ # Define an instance variable on this `MlflowClient` instance to reference the\n+ # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n+ # that the variable name is consistent with the variable name specified in the\n+ # preceding call to `getattr()`\n+ setattr(self, registry_client_attr, registry_client)\n+ except UnsupportedModelRegistryStoreURIException as exc:\n+ raise MlflowException(\n+ \"Model Registry features are not supported by the store with URI:\"\n+ f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n+ f\" {exc.supported_uri_schemes}.\",\n+ FEATURE_DISABLED,\n+ )\n+ return registry_client\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEventStr | WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of events that trigger this webhook. Can be strings or\n+ `WebhookEvent` objects.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self._get_registry_client().get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self._get_registry_client().list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEventStr | WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of events. Can be strings or `WebhookEvent` objects.\n+ secret: New webhook secret.\n+ status: New webhook status. Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ if events is not None:\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().update_webhook(\n+ webhook_id=webhook_id,\n+ name=name,\n+ description=description,\n+ url=url,\n+ events=events,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self._get_registry_client().delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n- return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,7 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,36 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n+++ a/mlflow/utils/credentials.py\n--- b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n+def get_default_host_creds(store_uri):\n+ creds = read_mlflow_creds()\n+ return MlflowHostCreds(\n+ host=store_uri,\n+ username=creds.username,\n+ password=creds.password,\n+ token=MLFLOW_TRACKING_TOKEN.get(),\n+ aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n+ auth=MLFLOW_TRACKING_AUTH.get(),\n+ ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n+ client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n+ server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n+def parse_dict(js_dict, message):\n+ \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n+ _stringify_all_experiment_ids(js_dict)\n+ ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n-\n-\n-\n-\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n+++ a/mlflow/utils/request_utils.py\n--- b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n+@lru_cache(maxsize=64)\n+def _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ # To create a new Session object for each process, we use the process id as the cache key.\n+ # This is to avoid sharing the same Session object across processes, which can lead to issues\n+ # such as https://stackoverflow.com/q/3724900.\n+ _pid,\n+ respect_retry_after_header=True,\n+):\n+ \"\"\"\n+ This function should not be called directly. Instead, use `_get_request_session` below.\n+ \"\"\"\n \n+ retry_kwargs = {\n+ \"total\": max_retries,\n+ \"connect\": max_retries,\n+ \"read\": max_retries,\n+ \"redirect\": max_retries,\n+ \"status\": max_retries,\n+ \"status_forcelist\": retry_codes,\n+ \"backoff_factor\": backoff_factor,\n+ \"backoff_jitter\": backoff_jitter,\n+ \"raise_on_status\": raise_on_status,\n+ \"respect_retry_after_header\": respect_retry_after_header,\n+ }\n+ urllib3_version = Version(urllib3.__version__)\n+ if urllib3_version >= Version(\"1.26.0\"):\n+ retry_kwargs[\"allowed_methods\"] = None\n+ else:\n+ retry_kwargs[\"method_whitelist\"] = None\n+\n+ if urllib3_version < Version(\"2.0\"):\n+ retry = JitteredRetry(**retry_kwargs)\n+ else:\n+ retry = Retry(**retry_kwargs)\n+ from mlflow.environment_variables import (\n+ MLFLOW_HTTP_POOL_CONNECTIONS,\n+ MLFLOW_HTTP_POOL_MAXSIZE,\n+ )\n+\n+ adapter = HTTPAdapter(\n+ pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n+ pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n+ max_retries=retry,\n+ )\n+ session = requests.Session()\n+ session.mount(\"https://\", adapter)\n+ session.mount(\"http://\", adapter)\n+ return session\n+\n+\n+def _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+):\n+ \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n+ Args:\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n+ Returns:\n+ requests.Session object.\n \n+ \"\"\"\n+ return _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ _pid=os.getpid(),\n+ respect_retry_after_header=respect_retry_after_header,\n+ )\n \n \n+def _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status=True,\n+ allow_redirects=None,\n+ respect_retry_after_header=True,\n+ **kwargs,\n+):\n+ \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n+ Args:\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ url: The target URL address for the HTTP request.\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ session = _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+ )\n \n+ # the environment variable is hardcoded here to avoid importing mlflow.\n+ # however, documentation is available in environment_variables.py\n+ env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n+ allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n+ return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n+from mlflow.utils.proto_json_utils import parse_dict\n+from mlflow.utils.request_utils import (\n+ _TRANSIENT_FAILURE_RESPONSE_CODES,\n+ _get_http_response_with_retries,\n+ augmented_raise_for_status, # noqa: F401\n+ cloud_storage_http_request, # noqa: F401\n+)\n+from mlflow.utils.string_utils import strip_suffix\n-\n-from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n-\n-\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n \n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n \n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n+\n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n \n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n+\n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n+\n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n+\n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n+\n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n+\n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n+\n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n+def _can_parse_as_json_object(string):\n+ try:\n+ return isinstance(json.loads(string), dict)\n+ except Exception:\n+ return False\n-\n-\n-\n-\n-\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n+def verify_rest_response(response, endpoint):\n+ \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n+ # Handle Armeria-specific response case where response text is \"200 OK\"\n+ # v1/traces endpoint might return empty response\n+ if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n+ response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n+ return response\n+\n+ # Handle non-200 status codes\n+ if response.status_code != 200:\n+ if _can_parse_as_json_object(response.text):\n+ raise RestException(json.loads(response.text))\n+ else:\n+ base_msg = (\n+ f\"API request to endpoint {endpoint} \"\n+ f\"failed with error code {response.status_code} != 200\"\n+ )\n+ raise MlflowException(\n+ f\"{base_msg}. Response body: '{response.text}'\",\n+ error_code=get_error_code(response.status_code),\n+ )\n \n+ # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n+ # response\n+ if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n+ base_msg = (\n+ \"API request to endpoint was successful but the response body was not \"\n+ \"in a valid JSON format\"\n+ )\n+ raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n+\n+ return response\n+\n+\n+def _validate_max_retries(max_retries):\n+ max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n+\n+ if max_retry_limit < 0:\n+ raise MlflowException(\n+ message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n+ \"Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if max_retries >= max_retry_limit:\n+ raise MlflowException(\n+ message=f\"The configured max_retries value ({max_retries}) is \"\n+ f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if max_retries < 0:\n+ raise MlflowException(\n+ message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+\n+def _validate_backoff_factor(backoff_factor):\n+ max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n+\n+ if max_backoff_factor_limit < 0:\n+ raise MlflowException(\n+ message=\"The current maximum backoff factor limit is invalid \"\n+ f\"({max_backoff_factor_limit}). Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor >= max_backoff_factor_limit:\n+ raise MlflowException(\n+ message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n+ \"of the maximum allowable backoff_factor limit \"\n+ f\"({max_backoff_factor_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor < 0:\n+ raise MlflowException(\n+ message=\"The backoff_factor value must be either 0 a positive integer. \"\n+ f\"Got {backoff_factor}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n+def call_endpoint(\n+ host_creds,\n+ endpoint,\n+ method,\n+ json_body,\n+ response_proto,\n+ extra_headers=None,\n+ retry_timeout_seconds=None,\n+):\n+ # Convert json string to json dictionary, to pass to requests\n+ if json_body is not None:\n+ json_body = json.loads(json_body)\n+ call_kwargs = {\n+ \"host_creds\": host_creds,\n+ \"endpoint\": endpoint,\n+ \"method\": method,\n+ }\n+ if extra_headers is not None:\n+ call_kwargs[\"extra_headers\"] = extra_headers\n+ if retry_timeout_seconds is not None:\n+ call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n+ if method == \"GET\":\n+ call_kwargs[\"params\"] = json_body\n+ response = http_request(**call_kwargs)\n+ else:\n+ call_kwargs[\"json\"] = json_body\n+ response = http_request(**call_kwargs)\n \n+ response = verify_rest_response(response, endpoint)\n+ response_to_parse = response.text\n+ try:\n+ js_dict = json.loads(response_to_parse)\n+ except json.JSONDecodeError:\n+ if len(response_to_parse) > 50:\n+ response_to_parse = response_to_parse[:50] + \"...\"\n+ _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n+ raise\n \n+ parse_dict(js_dict=js_dict, message=response_proto)\n+ return response_proto\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n+ def __init__(\n+ self,\n+ host,\n+ username=None,\n+ password=None,\n+ token=None,\n+ aws_sigv4=False,\n+ auth=None,\n+ ignore_tls_verification=False,\n+ client_cert_path=None,\n+ server_cert_path=None,\n+ use_databricks_sdk=False,\n+ databricks_auth_profile=None,\n+ client_id=None,\n+ client_secret=None,\n+ use_secret_scope_token=False,\n+ ):\n+ if not host:\n+ raise MlflowException(\n+ message=\"host is a required parameter for MlflowHostCreds\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if ignore_tls_verification and (server_cert_path is not None):\n+ raise MlflowException(\n+ message=(\n+ \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n+ \"must not be set! This error may have occurred because the \"\n+ \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n+ \"environment variables are both set - only one of these environment \"\n+ \"variables may be set.\"\n+ ),\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ self.host = host\n+ self.username = username\n+ self.password = password\n+ self.token = token\n+ self.aws_sigv4 = aws_sigv4\n+ self.auth = auth\n+ self.ignore_tls_verification = ignore_tls_verification\n+ self.client_cert_path = client_cert_path\n+ self.server_cert_path = server_cert_path\n+ self.use_databricks_sdk = use_databricks_sdk\n+ self.databricks_auth_profile = databricks_auth_profile\n+ self.client_id = client_id\n+ self.client_secret = client_secret\n+ self.use_secret_scope_token = use_secret_scope_token\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,11 +712,12 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n+ @property\n+ def verify(self):\n+ if self.use_databricks_sdk:\n+ # Let databricks-sdk set HTTP request `verify` param.\n+ return None\n+ if self.server_cert_path is None:\n+ return not self.ignore_tls_verification\n+ else:\n+ return self.server_cert_path\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..a7e2ebea0 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -833,6 +833,8 @@ class ModelRegistryClient:\n Returns:\n A :py:class:`mlflow.entities.webhook.Webhook` object.\n \"\"\"\n+ if not webhook_id:\n+ return None\n return self.store.get_webhook(webhook_id)\n \n @experimental(version=\"3.3.0\")\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py\ndeleted file mode 100644\nindex 97fd61b00..000000000\n--- a/tests/tracking/test_client_webhooks.py\n+++ /dev/null\n@@ -1,219 +0,0 @@\n-from pathlib import Path\n-from typing import Iterator\n-\n-import pytest\n-from cryptography.fernet import Fernet\n-\n-from mlflow.entities.webhook import WebhookAction, WebhookEntity, WebhookEvent, WebhookStatus\n-from mlflow.environment_variables import MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY\n-from mlflow.exceptions import MlflowException\n-from mlflow.server import handlers\n-from mlflow.server.fastapi_app import app\n-from mlflow.server.handlers import initialize_backend_stores\n-from mlflow.tracking import MlflowClient\n-\n-from tests.helper_functions import get_safe_port\n-from tests.tracking.integration_test_utils import ServerThread\n-\n-\n-@pytest.fixture\n-def client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Iterator[MlflowClient]:\n- \"\"\"Setup a local MLflow server with proper webhook encryption key support.\"\"\"\n- # Set up encryption key for webhooks using monkeypatch\n- encryption_key = Fernet.generate_key().decode(\"utf-8\")\n- monkeypatch.setenv(MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY.name, encryption_key)\n-\n- # Configure backend stores\n- backend_uri = f\"sqlite:///{tmp_path / 'mlflow.db'}\"\n- default_artifact_root = tmp_path.as_uri()\n-\n- # Force-reset backend stores before each test\n- handlers._tracking_store = None\n- handlers._model_registry_store = None\n- initialize_backend_stores(backend_uri, default_artifact_root=default_artifact_root)\n-\n- # Start server and return client\n- with ServerThread(app, get_safe_port()) as url:\n- yield MlflowClient(url)\n-\n-\n-def test_create_webhook(client: MlflowClient):\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n- assert webhook.events == [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n-\n- webhook = client.get_webhook(webhook.webhook_id)\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n-\n- # With secret\n- webhook_with_secret = client.create_webhook(\n- name=\"test_webhook_with_secret\",\n- url=\"https://example.com/webhook_with_secret\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- secret=\"my_secret\",\n- )\n- assert webhook_with_secret.name == \"test_webhook_with_secret\"\n- assert webhook_with_secret.url == \"https://example.com/webhook_with_secret\"\n- assert webhook_with_secret.secret is None\n- assert webhook_with_secret.events == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)\n- ]\n-\n- # Multiple events\n- webhook_multiple_events = client.create_webhook(\n- name=\"test_webhook_multiple_events\",\n- url=\"https://example.com/webhook_multiple_events\",\n- events=[\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- ],\n- )\n- assert webhook_multiple_events.name == \"test_webhook_multiple_events\"\n- assert webhook_multiple_events.url == \"https://example.com/webhook_multiple_events\"\n- assert sorted(\n- webhook_multiple_events.events, key=lambda e: (e.entity.value, e.action.value)\n- ) == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- ]\n- assert webhook_multiple_events.secret is None\n-\n-\n-def test_get_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- created_webhook = client.create_webhook(\n- name=\"test_webhook\", url=\"https://example.com/webhook\", events=events\n- )\n- retrieved_webhook = client.get_webhook(created_webhook.webhook_id)\n- assert retrieved_webhook.webhook_id == created_webhook.webhook_id\n- assert retrieved_webhook.name == \"test_webhook\"\n- assert retrieved_webhook.url == \"https://example.com/webhook\"\n- assert retrieved_webhook.events == events\n-\n-\n-def test_get_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.get_webhook(\"nonexistent\")\n-\n-\n-def test_list_webhooks(client: MlflowClient):\n- # Create more webhooks than max_results\n- created_webhooks = []\n- for i in range(5):\n- webhook = client.create_webhook(\n- name=f\"webhook{i}\",\n- url=f\"https://example.com/{i}\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- created_webhooks.append(webhook)\n- # Test pagination with max_results=2\n- webhooks_page = client.list_webhooks(max_results=2)\n- assert len(webhooks_page) == 2\n- assert webhooks_page.token is not None\n- # Get next page\n- next_webhooks_page = client.list_webhooks(max_results=2, page_token=webhooks_page.token)\n- assert len(next_webhooks_page) == 2\n- assert next_webhooks_page.token is not None\n- # Verify we don't get duplicates\n- first_page_ids = {w.webhook_id for w in webhooks_page}\n- second_page_ids = {w.webhook_id for w in next_webhooks_page}\n- assert first_page_ids.isdisjoint(second_page_ids)\n-\n-\n-def test_update_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\", url=\"https://example.com/original\", events=events\n- )\n- # Update webhook\n- new_events = [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.REGISTERED_MODEL, WebhookAction.CREATED),\n- ]\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- url=\"https://example.com/updated\",\n- events=new_events,\n- description=\"Updated description\",\n- secret=\"new_secret\",\n- status=WebhookStatus.DISABLED,\n- )\n- assert updated_webhook.webhook_id == webhook.webhook_id\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/updated\"\n- assert updated_webhook.events == new_events\n- assert updated_webhook.description == \"Updated description\"\n- assert updated_webhook.status == WebhookStatus.DISABLED\n- assert updated_webhook.last_updated_timestamp > webhook.last_updated_timestamp\n-\n-\n-def test_update_webhook_partial(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\",\n- url=\"https://example.com/original\",\n- events=events,\n- description=\"Original description\",\n- )\n- # Update only the name\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- )\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/original\"\n- assert updated_webhook.events == events\n- assert updated_webhook.description == \"Original description\"\n-\n-\n-def test_update_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.update_webhook(webhook_id=\"nonexistent\", name=\"new_name\")\n-\n-\n-@pytest.mark.parametrize(\n- (\"invalid_url\", \"expected_match\"),\n- [\n- (\" \", r\"Webhook URL cannot be empty or just whitespace\"),\n- (\"ftp://example.com\", r\"Invalid webhook URL scheme\"),\n- (\"http://[invalid\", r\"Invalid webhook URL\"),\n- ],\n-)\n-def test_update_webhook_invalid_urls(client: MlflowClient, invalid_url: str, expected_match: str):\n- # Create a valid webhook first\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- with pytest.raises(MlflowException, match=expected_match):\n- client.update_webhook(webhook_id=webhook.webhook_id, url=invalid_url)\n-\n-\n-def test_delete_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=events,\n- )\n- client.delete_webhook(webhook.webhook_id)\n- with pytest.raises(MlflowException, match=r\"Webhook with ID .* not found\"):\n- client.get_webhook(webhook.webhook_id)\n- webhooks_page = client.list_webhooks()\n- webhook_ids = {w.webhook_id for w in webhooks_page}\n- assert webhook.webhook_id not in webhook_ids\n-\n-\n-def test_delete_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.delete_webhook(\"nonexistent\")\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S5\",\n \"shared_functions\": 2,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":58,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S7\",\n \"text\": \"List webhooks. Signature: list_webhooks(self, max_results: int | None = None, page_token: str | None = None) -> PagedList[Webhook]. Args: max_results: Maximum number of webhooks to return. page_token: Token specifying the next page of results. Returns: A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects. Task statement: Support pagination for webhook listing operations.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracking/client.py\",\n \"lines\": \"6368-6383\",\n \"function\": \"MlflowClient.list_webhooks\"\n },\n {\n \"file\": \"mlflow/tracking/_model_registry/client.py\",\n \"lines\": \"838-854\",\n \"function\": \"ModelRegistryClient.list_webhooks\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n--- a/mlflow/entities/webhook.py\n+++ b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n- def __eq__(self, other: object) -> bool:\n- if not isinstance(other, WebhookEvent):\n- return False\n- return self.entity == other.entity and self.action == other.action\n+\n+\n+\n+\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n- @property\n- def secret(self) -> str | None:\n- return self._secret\n+\n+\n+\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n--- a/mlflow/server/handlers.py\n+++ b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n--- a/mlflow/store/artifact/azure_blob_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-from mlflow.utils.credentials import get_default_host_creds\n+\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- augmented_raise_for_status,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n- def __init__(self, store_uri=None, tracking_uri=None):\n- \"\"\"\n- Empty constructor. This is deliberately not marked as abstract, else every derived class\n- would be forced to create one.\n \n- Args:\n- store_uri: The model registry store URI.\n- tracking_uri: URI of the current MLflow tracking server, used to perform operations\n- like fetching source run metadata or downloading source run artifacts\n- to support subsequently uploading them to the model registry storage\n- location.\n- \"\"\"\n- # Create a thread lock to ensure thread safety when linking prompts to other entities,\n- # since the default linking implementation reads and appends entity tags, which\n- # is prone to concurrent modification issues\n- self._prompt_link_lock = threading.RLock()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n--- a/mlflow/store/model_registry/base_rest_store.py\n+++ b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n-from mlflow.utils.rest_utils import (\n- call_endpoint,\n- call_endpoints,\n-)\n+from mlflow.utils.rest_utils import call_endpoints\n+\n+\n+\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- @record_usage_event(CreateWebhookEvent)\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Unique name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of event types that trigger this webhook.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE).\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- return self.store.create_webhook(name, url, events, description, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self.store.get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self.store.list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of event types.\n- secret: New webhook secret.\n- status: New webhook status.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self.store.delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n- return self.store.test_webhook(webhook_id, event)\n+ return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n--- a/mlflow/tracking/_model_registry/registry.py\n+++ b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n- from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n+\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n--- a/mlflow/tracking/_model_registry/utils.py\n+++ b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n-def _get_registry_uri_from_spark_session():\n- session = _get_active_spark_session()\n- if session is None:\n- return None\n \n- if is_in_databricks_serverless_runtime():\n- # Connected to Serverless\n- return \"databricks-uc\"\n \n- from pyspark.sql.utils import AnalysisException\n \n- try:\n- return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n- except AnalysisException:\n- # In serverless clusters, session.conf.get() is unsupported\n- # and raises an AnalysisException. We may encounter this case\n- # when DBConnect is used to connect to a serverless cluster,\n- # in which case the prior `is_in_databricks_serverless_runtime()`\n- # check will have returned false (as of 2025-06-07, it checks\n- # an environment variable that isn't set by DBConnect)\n- return None\n \n \n-def _get_registry_uri_from_context():\n- if _registry_uri is not None:\n- return _registry_uri\n- elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n- return uri\n- return _registry_uri\n \n \n-def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n- \"\"\"\n- Get the default registry URI for a given tracking URI.\n \n- If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n- Otherwise, returns the tracking URI itself.\n \n- Args:\n- tracking_uri: The tracking URI to get the default registry URI for\n \n- Returns:\n- The default registry URI\n- \"\"\"\n- if tracking_uri is not None and is_databricks_uri(tracking_uri):\n- # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n- # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n- # model registry offering on Databricks\n- if tracking_uri == \"databricks\":\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n- else:\n- # Extract profile from tracking URI and construct databricks-uc URI\n- profile, key_prefix = get_db_info_from_uri(tracking_uri)\n- if profile:\n- # Reconstruct the profile string including key_prefix if present\n- profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n- return construct_db_uc_uri_from_profile(profile_string)\n- else:\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n-\n- # For non-databricks tracking URIs, use the tracking URI as the registry URI\n- return tracking_uri\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n-def _resolve_registry_uri(\n- registry_uri: str | None = None, tracking_uri: str | None = None\n-) -> str | None:\n- \"\"\"\n- Resolve the registry URI following the same logic as get_registry_uri().\n- \"\"\"\n- return (\n- registry_uri\n- or _get_registry_uri_from_context()\n- or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n-def _get_rest_store(store_uri, **_):\n- return RestStore(partial(get_default_host_creds, store_uri))\n+\n+\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n-def _get_store_registry():\n- global _model_registry_store_registry\n- from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n- from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n- if _model_registry_store_registry is not None:\n- return _model_registry_store_registry\n \n- _model_registry_store_registry = ModelRegistryStoreRegistry()\n- _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n- # Register a placeholder function that raises if users pass a registry URI with scheme\n- # \"databricks-uc\"\n- _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n- _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n- for scheme in [\"http\", \"https\"]:\n- _model_registry_store_registry.register(scheme, _get_rest_store)\n \n- for scheme in DATABASE_ENGINES:\n- _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n- for scheme in [\"\", \"file\"]:\n- _model_registry_store_registry.register(scheme, _get_file_store)\n \n- _model_registry_store_registry.register_entrypoints()\n- return _model_registry_store_registry\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n--- a/mlflow/tracking/_tracking_service/utils.py\n+++ b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n- @property\n- def tracking_uri(self):\n- return self._tracking_client.tracking_uri\n \n- def _get_registry_client(self):\n- \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n- Raises:\n- MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n- example, when the registry URI refers to an unsupported store type (e.g., the\n- FileStore).\n \n- Returns:\n- A ModelRegistryClient instance.\n- \"\"\"\n- # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n- # an instance variable on this `MlflowClient` instance. Because the instance variable\n- # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n- # function is used to safely fetch the variable (if it is defined) or a NoneType\n- # (if it is not defined)\n- registry_client_attr = \"_registry_client_lazy\"\n- registry_client = getattr(self, registry_client_attr, None)\n- if registry_client is None:\n- try:\n- registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n- # Define an instance variable on this `MlflowClient` instance to reference the\n- # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n- # that the variable name is consistent with the variable name specified in the\n- # preceding call to `getattr()`\n- setattr(self, registry_client_attr, registry_client)\n- except UnsupportedModelRegistryStoreURIException as exc:\n- raise MlflowException(\n- \"Model Registry features are not supported by the store with URI:\"\n- f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n- f\" {exc.supported_uri_schemes}.\",\n- FEATURE_DISABLED,\n- )\n- return registry_client\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEventStr | WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of events that trigger this webhook. Can be strings or\n- `WebhookEvent` objects.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self._get_registry_client().get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self._get_registry_client().list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEventStr | WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of events. Can be strings or `WebhookEvent` objects.\n- secret: New webhook secret.\n- status: New webhook status. Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().update_webhook(\n- webhook_id=webhook_id,\n- name=name,\n- description=description,\n- url=url,\n- events=events,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self._get_registry_client().delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n- return self._get_registry_client().test_webhook(webhook_id, event)\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,7 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,36 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n--- a/mlflow/utils/credentials.py\n+++ b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n-def get_default_host_creds(store_uri):\n- creds = read_mlflow_creds()\n- return MlflowHostCreds(\n- host=store_uri,\n- username=creds.username,\n- password=creds.password,\n- token=MLFLOW_TRACKING_TOKEN.get(),\n- aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n- auth=MLFLOW_TRACKING_AUTH.get(),\n- ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n- client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n- server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n-def parse_dict(js_dict, message):\n- \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n- _stringify_all_experiment_ids(js_dict)\n- ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n+\n+\n+\n+\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n--- a/mlflow/utils/request_utils.py\n+++ b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n-@lru_cache(maxsize=64)\n-def _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- # To create a new Session object for each process, we use the process id as the cache key.\n- # This is to avoid sharing the same Session object across processes, which can lead to issues\n- # such as https://stackoverflow.com/q/3724900.\n- _pid,\n- respect_retry_after_header=True,\n-):\n- \"\"\"\n- This function should not be called directly. Instead, use `_get_request_session` below.\n- \"\"\"\n \n- retry_kwargs = {\n- \"total\": max_retries,\n- \"connect\": max_retries,\n- \"read\": max_retries,\n- \"redirect\": max_retries,\n- \"status\": max_retries,\n- \"status_forcelist\": retry_codes,\n- \"backoff_factor\": backoff_factor,\n- \"backoff_jitter\": backoff_jitter,\n- \"raise_on_status\": raise_on_status,\n- \"respect_retry_after_header\": respect_retry_after_header,\n- }\n- urllib3_version = Version(urllib3.__version__)\n- if urllib3_version >= Version(\"1.26.0\"):\n- retry_kwargs[\"allowed_methods\"] = None\n- else:\n- retry_kwargs[\"method_whitelist\"] = None\n-\n- if urllib3_version < Version(\"2.0\"):\n- retry = JitteredRetry(**retry_kwargs)\n- else:\n- retry = Retry(**retry_kwargs)\n- from mlflow.environment_variables import (\n- MLFLOW_HTTP_POOL_CONNECTIONS,\n- MLFLOW_HTTP_POOL_MAXSIZE,\n- )\n-\n- adapter = HTTPAdapter(\n- pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n- pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n- max_retries=retry,\n- )\n- session = requests.Session()\n- session.mount(\"https://\", adapter)\n- session.mount(\"http://\", adapter)\n- return session\n-\n-\n-def _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n-):\n- \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n- Args:\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n- Returns:\n- requests.Session object.\n \n- \"\"\"\n- return _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- _pid=os.getpid(),\n- respect_retry_after_header=respect_retry_after_header,\n- )\n \n \n-def _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status=True,\n- allow_redirects=None,\n- respect_retry_after_header=True,\n- **kwargs,\n-):\n- \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n- Args:\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- url: The target URL address for the HTTP request.\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- session = _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n- )\n \n- # the environment variable is hardcoded here to avoid importing mlflow.\n- # however, documentation is available in environment_variables.py\n- env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n- allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n- return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n-from mlflow.utils.proto_json_utils import parse_dict\n-from mlflow.utils.request_utils import (\n- _TRANSIENT_FAILURE_RESPONSE_CODES,\n- _get_http_response_with_retries,\n- augmented_raise_for_status, # noqa: F401\n- cloud_storage_http_request, # noqa: F401\n-)\n-from mlflow.utils.string_utils import strip_suffix\n+\n+from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n+\n+\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n \n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n \n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n-\n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n \n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n-\n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n-\n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n-\n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n-\n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n-\n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n-\n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n-def _can_parse_as_json_object(string):\n- try:\n- return isinstance(json.loads(string), dict)\n- except Exception:\n- return False\n+\n+\n+\n+\n+\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n-def verify_rest_response(response, endpoint):\n- \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n- # Handle Armeria-specific response case where response text is \"200 OK\"\n- # v1/traces endpoint might return empty response\n- if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n- response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n- return response\n-\n- # Handle non-200 status codes\n- if response.status_code != 200:\n- if _can_parse_as_json_object(response.text):\n- raise RestException(json.loads(response.text))\n- else:\n- base_msg = (\n- f\"API request to endpoint {endpoint} \"\n- f\"failed with error code {response.status_code} != 200\"\n- )\n- raise MlflowException(\n- f\"{base_msg}. Response body: '{response.text}'\",\n- error_code=get_error_code(response.status_code),\n- )\n \n- # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n- # response\n- if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n- base_msg = (\n- \"API request to endpoint was successful but the response body was not \"\n- \"in a valid JSON format\"\n- )\n- raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n-\n- return response\n-\n-\n-def _validate_max_retries(max_retries):\n- max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n-\n- if max_retry_limit < 0:\n- raise MlflowException(\n- message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n- \"Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if max_retries >= max_retry_limit:\n- raise MlflowException(\n- message=f\"The configured max_retries value ({max_retries}) is \"\n- f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if max_retries < 0:\n- raise MlflowException(\n- message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n-\n-def _validate_backoff_factor(backoff_factor):\n- max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n-\n- if max_backoff_factor_limit < 0:\n- raise MlflowException(\n- message=\"The current maximum backoff factor limit is invalid \"\n- f\"({max_backoff_factor_limit}). Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor >= max_backoff_factor_limit:\n- raise MlflowException(\n- message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n- \"of the maximum allowable backoff_factor limit \"\n- f\"({max_backoff_factor_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor < 0:\n- raise MlflowException(\n- message=\"The backoff_factor value must be either 0 a positive integer. \"\n- f\"Got {backoff_factor}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n-def call_endpoint(\n- host_creds,\n- endpoint,\n- method,\n- json_body,\n- response_proto,\n- extra_headers=None,\n- retry_timeout_seconds=None,\n-):\n- # Convert json string to json dictionary, to pass to requests\n- if json_body is not None:\n- json_body = json.loads(json_body)\n- call_kwargs = {\n- \"host_creds\": host_creds,\n- \"endpoint\": endpoint,\n- \"method\": method,\n- }\n- if extra_headers is not None:\n- call_kwargs[\"extra_headers\"] = extra_headers\n- if retry_timeout_seconds is not None:\n- call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n- if method == \"GET\":\n- call_kwargs[\"params\"] = json_body\n- response = http_request(**call_kwargs)\n- else:\n- call_kwargs[\"json\"] = json_body\n- response = http_request(**call_kwargs)\n \n- response = verify_rest_response(response, endpoint)\n- response_to_parse = response.text\n- try:\n- js_dict = json.loads(response_to_parse)\n- except json.JSONDecodeError:\n- if len(response_to_parse) > 50:\n- response_to_parse = response_to_parse[:50] + \"...\"\n- _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n- raise\n \n- parse_dict(js_dict=js_dict, message=response_proto)\n- return response_proto\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n- def __init__(\n- self,\n- host,\n- username=None,\n- password=None,\n- token=None,\n- aws_sigv4=False,\n- auth=None,\n- ignore_tls_verification=False,\n- client_cert_path=None,\n- server_cert_path=None,\n- use_databricks_sdk=False,\n- databricks_auth_profile=None,\n- client_id=None,\n- client_secret=None,\n- use_secret_scope_token=False,\n- ):\n- if not host:\n- raise MlflowException(\n- message=\"host is a required parameter for MlflowHostCreds\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if ignore_tls_verification and (server_cert_path is not None):\n- raise MlflowException(\n- message=(\n- \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n- \"must not be set! This error may have occurred because the \"\n- \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n- \"environment variables are both set - only one of these environment \"\n- \"variables may be set.\"\n- ),\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- self.host = host\n- self.username = username\n- self.password = password\n- self.token = token\n- self.aws_sigv4 = aws_sigv4\n- self.auth = auth\n- self.ignore_tls_verification = ignore_tls_verification\n- self.client_cert_path = client_cert_path\n- self.server_cert_path = server_cert_path\n- self.use_databricks_sdk = use_databricks_sdk\n- self.databricks_auth_profile = databricks_auth_profile\n- self.client_id = client_id\n- self.client_secret = client_secret\n- self.use_secret_scope_token = use_secret_scope_token\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,12 +712,11 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n- @property\n- def verify(self):\n- if self.use_databricks_sdk:\n- # Let databricks-sdk set HTTP request `verify` param.\n- return None\n- if self.server_cert_path is None:\n- return not self.ignore_tls_verification\n- else:\n- return self.server_cert_path\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n+++ a/mlflow/entities/webhook.py\n--- b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n+ def __eq__(self, other: object) -> bool:\n+ if not isinstance(other, WebhookEvent):\n+ return False\n+ return self.entity == other.entity and self.action == other.action\n-\n-\n-\n-\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n+ @property\n+ def secret(self) -> str | None:\n+ return self._secret\n-\n-\n-\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n+++ a/mlflow/server/handlers.py\n--- b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n+++ a/mlflow/store/artifact/azure_blob_artifact_repo.py\n--- b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+from mlflow.utils.credentials import get_default_host_creds\n-\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ augmented_raise_for_status,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(self, store_uri=None, tracking_uri=None):\n+ \"\"\"\n+ Empty constructor. This is deliberately not marked as abstract, else every derived class\n+ would be forced to create one.\n \n+ Args:\n+ store_uri: The model registry store URI.\n+ tracking_uri: URI of the current MLflow tracking server, used to perform operations\n+ like fetching source run metadata or downloading source run artifacts\n+ to support subsequently uploading them to the model registry storage\n+ location.\n+ \"\"\"\n+ # Create a thread lock to ensure thread safety when linking prompts to other entities,\n+ # since the default linking implementation reads and appends entity tags, which\n+ # is prone to concurrent modification issues\n+ self._prompt_link_lock = threading.RLock()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n+++ a/mlflow/store/model_registry/base_rest_store.py\n--- b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n+from mlflow.utils.rest_utils import (\n+ call_endpoint,\n+ call_endpoints,\n+)\n-from mlflow.utils.rest_utils import call_endpoints\n-\n-\n-\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n+++ a/mlflow/tracking/_model_registry/client.py\n--- b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ @record_usage_event(CreateWebhookEvent)\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Unique name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of event types that trigger this webhook.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE).\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ return self.store.create_webhook(name, url, events, description, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self.store.get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self.store.list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of event types.\n+ secret: New webhook secret.\n+ status: New webhook status.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self.store.delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n+ return self.store.test_webhook(webhook_id, event)\n- return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n+++ a/mlflow/tracking/_model_registry/registry.py\n--- b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n+ from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n-\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n+++ a/mlflow/tracking/_model_registry/utils.py\n--- b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n+def _get_registry_uri_from_spark_session():\n+ session = _get_active_spark_session()\n+ if session is None:\n+ return None\n \n+ if is_in_databricks_serverless_runtime():\n+ # Connected to Serverless\n+ return \"databricks-uc\"\n \n+ from pyspark.sql.utils import AnalysisException\n \n+ try:\n+ return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n+ except AnalysisException:\n+ # In serverless clusters, session.conf.get() is unsupported\n+ # and raises an AnalysisException. We may encounter this case\n+ # when DBConnect is used to connect to a serverless cluster,\n+ # in which case the prior `is_in_databricks_serverless_runtime()`\n+ # check will have returned false (as of 2025-06-07, it checks\n+ # an environment variable that isn't set by DBConnect)\n+ return None\n \n \n+def _get_registry_uri_from_context():\n+ if _registry_uri is not None:\n+ return _registry_uri\n+ elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n+ return uri\n+ return _registry_uri\n \n \n+def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n+ \"\"\"\n+ Get the default registry URI for a given tracking URI.\n \n+ If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n+ Otherwise, returns the tracking URI itself.\n \n+ Args:\n+ tracking_uri: The tracking URI to get the default registry URI for\n \n+ Returns:\n+ The default registry URI\n+ \"\"\"\n+ if tracking_uri is not None and is_databricks_uri(tracking_uri):\n+ # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n+ # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n+ # model registry offering on Databricks\n+ if tracking_uri == \"databricks\":\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+ else:\n+ # Extract profile from tracking URI and construct databricks-uc URI\n+ profile, key_prefix = get_db_info_from_uri(tracking_uri)\n+ if profile:\n+ # Reconstruct the profile string including key_prefix if present\n+ profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n+ return construct_db_uc_uri_from_profile(profile_string)\n+ else:\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+\n+ # For non-databricks tracking URIs, use the tracking URI as the registry URI\n+ return tracking_uri\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n+def _resolve_registry_uri(\n+ registry_uri: str | None = None, tracking_uri: str | None = None\n+) -> str | None:\n+ \"\"\"\n+ Resolve the registry URI following the same logic as get_registry_uri().\n+ \"\"\"\n+ return (\n+ registry_uri\n+ or _get_registry_uri_from_context()\n+ or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n+def _get_rest_store(store_uri, **_):\n+ return RestStore(partial(get_default_host_creds, store_uri))\n-\n-\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n+def _get_store_registry():\n+ global _model_registry_store_registry\n+ from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n+ from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n+ if _model_registry_store_registry is not None:\n+ return _model_registry_store_registry\n \n+ _model_registry_store_registry = ModelRegistryStoreRegistry()\n+ _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n+ # Register a placeholder function that raises if users pass a registry URI with scheme\n+ # \"databricks-uc\"\n+ _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n+ _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n+ for scheme in [\"http\", \"https\"]:\n+ _model_registry_store_registry.register(scheme, _get_rest_store)\n \n+ for scheme in DATABASE_ENGINES:\n+ _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n+ for scheme in [\"\", \"file\"]:\n+ _model_registry_store_registry.register(scheme, _get_file_store)\n \n+ _model_registry_store_registry.register_entrypoints()\n+ return _model_registry_store_registry\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n+++ a/mlflow/tracking/_tracking_service/utils.py\n--- b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n+++ a/mlflow/tracking/client.py\n--- b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n+ @property\n+ def tracking_uri(self):\n+ return self._tracking_client.tracking_uri\n \n+ def _get_registry_client(self):\n+ \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n+ Raises:\n+ MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n+ example, when the registry URI refers to an unsupported store type (e.g., the\n+ FileStore).\n \n+ Returns:\n+ A ModelRegistryClient instance.\n+ \"\"\"\n+ # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n+ # an instance variable on this `MlflowClient` instance. Because the instance variable\n+ # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n+ # function is used to safely fetch the variable (if it is defined) or a NoneType\n+ # (if it is not defined)\n+ registry_client_attr = \"_registry_client_lazy\"\n+ registry_client = getattr(self, registry_client_attr, None)\n+ if registry_client is None:\n+ try:\n+ registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n+ # Define an instance variable on this `MlflowClient` instance to reference the\n+ # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n+ # that the variable name is consistent with the variable name specified in the\n+ # preceding call to `getattr()`\n+ setattr(self, registry_client_attr, registry_client)\n+ except UnsupportedModelRegistryStoreURIException as exc:\n+ raise MlflowException(\n+ \"Model Registry features are not supported by the store with URI:\"\n+ f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n+ f\" {exc.supported_uri_schemes}.\",\n+ FEATURE_DISABLED,\n+ )\n+ return registry_client\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEventStr | WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of events that trigger this webhook. Can be strings or\n+ `WebhookEvent` objects.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self._get_registry_client().get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self._get_registry_client().list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEventStr | WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of events. Can be strings or `WebhookEvent` objects.\n+ secret: New webhook secret.\n+ status: New webhook status. Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ if events is not None:\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().update_webhook(\n+ webhook_id=webhook_id,\n+ name=name,\n+ description=description,\n+ url=url,\n+ events=events,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self._get_registry_client().delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n- return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,7 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,36 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n+++ a/mlflow/utils/credentials.py\n--- b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n+def get_default_host_creds(store_uri):\n+ creds = read_mlflow_creds()\n+ return MlflowHostCreds(\n+ host=store_uri,\n+ username=creds.username,\n+ password=creds.password,\n+ token=MLFLOW_TRACKING_TOKEN.get(),\n+ aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n+ auth=MLFLOW_TRACKING_AUTH.get(),\n+ ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n+ client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n+ server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n+def parse_dict(js_dict, message):\n+ \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n+ _stringify_all_experiment_ids(js_dict)\n+ ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n-\n-\n-\n-\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n+++ a/mlflow/utils/request_utils.py\n--- b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n+@lru_cache(maxsize=64)\n+def _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ # To create a new Session object for each process, we use the process id as the cache key.\n+ # This is to avoid sharing the same Session object across processes, which can lead to issues\n+ # such as https://stackoverflow.com/q/3724900.\n+ _pid,\n+ respect_retry_after_header=True,\n+):\n+ \"\"\"\n+ This function should not be called directly. Instead, use `_get_request_session` below.\n+ \"\"\"\n \n+ retry_kwargs = {\n+ \"total\": max_retries,\n+ \"connect\": max_retries,\n+ \"read\": max_retries,\n+ \"redirect\": max_retries,\n+ \"status\": max_retries,\n+ \"status_forcelist\": retry_codes,\n+ \"backoff_factor\": backoff_factor,\n+ \"backoff_jitter\": backoff_jitter,\n+ \"raise_on_status\": raise_on_status,\n+ \"respect_retry_after_header\": respect_retry_after_header,\n+ }\n+ urllib3_version = Version(urllib3.__version__)\n+ if urllib3_version >= Version(\"1.26.0\"):\n+ retry_kwargs[\"allowed_methods\"] = None\n+ else:\n+ retry_kwargs[\"method_whitelist\"] = None\n+\n+ if urllib3_version < Version(\"2.0\"):\n+ retry = JitteredRetry(**retry_kwargs)\n+ else:\n+ retry = Retry(**retry_kwargs)\n+ from mlflow.environment_variables import (\n+ MLFLOW_HTTP_POOL_CONNECTIONS,\n+ MLFLOW_HTTP_POOL_MAXSIZE,\n+ )\n+\n+ adapter = HTTPAdapter(\n+ pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n+ pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n+ max_retries=retry,\n+ )\n+ session = requests.Session()\n+ session.mount(\"https://\", adapter)\n+ session.mount(\"http://\", adapter)\n+ return session\n+\n+\n+def _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+):\n+ \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n+ Args:\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n+ Returns:\n+ requests.Session object.\n \n+ \"\"\"\n+ return _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ _pid=os.getpid(),\n+ respect_retry_after_header=respect_retry_after_header,\n+ )\n \n \n+def _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status=True,\n+ allow_redirects=None,\n+ respect_retry_after_header=True,\n+ **kwargs,\n+):\n+ \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n+ Args:\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ url: The target URL address for the HTTP request.\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ session = _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+ )\n \n+ # the environment variable is hardcoded here to avoid importing mlflow.\n+ # however, documentation is available in environment_variables.py\n+ env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n+ allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n+ return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n+from mlflow.utils.proto_json_utils import parse_dict\n+from mlflow.utils.request_utils import (\n+ _TRANSIENT_FAILURE_RESPONSE_CODES,\n+ _get_http_response_with_retries,\n+ augmented_raise_for_status, # noqa: F401\n+ cloud_storage_http_request, # noqa: F401\n+)\n+from mlflow.utils.string_utils import strip_suffix\n-\n-from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n-\n-\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n \n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n \n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n+\n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n \n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n+\n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n+\n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n+\n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n+\n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n+\n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n+\n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n+def _can_parse_as_json_object(string):\n+ try:\n+ return isinstance(json.loads(string), dict)\n+ except Exception:\n+ return False\n-\n-\n-\n-\n-\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n+def verify_rest_response(response, endpoint):\n+ \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n+ # Handle Armeria-specific response case where response text is \"200 OK\"\n+ # v1/traces endpoint might return empty response\n+ if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n+ response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n+ return response\n+\n+ # Handle non-200 status codes\n+ if response.status_code != 200:\n+ if _can_parse_as_json_object(response.text):\n+ raise RestException(json.loads(response.text))\n+ else:\n+ base_msg = (\n+ f\"API request to endpoint {endpoint} \"\n+ f\"failed with error code {response.status_code} != 200\"\n+ )\n+ raise MlflowException(\n+ f\"{base_msg}. Response body: '{response.text}'\",\n+ error_code=get_error_code(response.status_code),\n+ )\n \n+ # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n+ # response\n+ if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n+ base_msg = (\n+ \"API request to endpoint was successful but the response body was not \"\n+ \"in a valid JSON format\"\n+ )\n+ raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n+\n+ return response\n+\n+\n+def _validate_max_retries(max_retries):\n+ max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n+\n+ if max_retry_limit < 0:\n+ raise MlflowException(\n+ message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n+ \"Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if max_retries >= max_retry_limit:\n+ raise MlflowException(\n+ message=f\"The configured max_retries value ({max_retries}) is \"\n+ f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if max_retries < 0:\n+ raise MlflowException(\n+ message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+\n+def _validate_backoff_factor(backoff_factor):\n+ max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n+\n+ if max_backoff_factor_limit < 0:\n+ raise MlflowException(\n+ message=\"The current maximum backoff factor limit is invalid \"\n+ f\"({max_backoff_factor_limit}). Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor >= max_backoff_factor_limit:\n+ raise MlflowException(\n+ message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n+ \"of the maximum allowable backoff_factor limit \"\n+ f\"({max_backoff_factor_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor < 0:\n+ raise MlflowException(\n+ message=\"The backoff_factor value must be either 0 a positive integer. \"\n+ f\"Got {backoff_factor}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n+def call_endpoint(\n+ host_creds,\n+ endpoint,\n+ method,\n+ json_body,\n+ response_proto,\n+ extra_headers=None,\n+ retry_timeout_seconds=None,\n+):\n+ # Convert json string to json dictionary, to pass to requests\n+ if json_body is not None:\n+ json_body = json.loads(json_body)\n+ call_kwargs = {\n+ \"host_creds\": host_creds,\n+ \"endpoint\": endpoint,\n+ \"method\": method,\n+ }\n+ if extra_headers is not None:\n+ call_kwargs[\"extra_headers\"] = extra_headers\n+ if retry_timeout_seconds is not None:\n+ call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n+ if method == \"GET\":\n+ call_kwargs[\"params\"] = json_body\n+ response = http_request(**call_kwargs)\n+ else:\n+ call_kwargs[\"json\"] = json_body\n+ response = http_request(**call_kwargs)\n \n+ response = verify_rest_response(response, endpoint)\n+ response_to_parse = response.text\n+ try:\n+ js_dict = json.loads(response_to_parse)\n+ except json.JSONDecodeError:\n+ if len(response_to_parse) > 50:\n+ response_to_parse = response_to_parse[:50] + \"...\"\n+ _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n+ raise\n \n+ parse_dict(js_dict=js_dict, message=response_proto)\n+ return response_proto\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n+ def __init__(\n+ self,\n+ host,\n+ username=None,\n+ password=None,\n+ token=None,\n+ aws_sigv4=False,\n+ auth=None,\n+ ignore_tls_verification=False,\n+ client_cert_path=None,\n+ server_cert_path=None,\n+ use_databricks_sdk=False,\n+ databricks_auth_profile=None,\n+ client_id=None,\n+ client_secret=None,\n+ use_secret_scope_token=False,\n+ ):\n+ if not host:\n+ raise MlflowException(\n+ message=\"host is a required parameter for MlflowHostCreds\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if ignore_tls_verification and (server_cert_path is not None):\n+ raise MlflowException(\n+ message=(\n+ \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n+ \"must not be set! This error may have occurred because the \"\n+ \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n+ \"environment variables are both set - only one of these environment \"\n+ \"variables may be set.\"\n+ ),\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ self.host = host\n+ self.username = username\n+ self.password = password\n+ self.token = token\n+ self.aws_sigv4 = aws_sigv4\n+ self.auth = auth\n+ self.ignore_tls_verification = ignore_tls_verification\n+ self.client_cert_path = client_cert_path\n+ self.server_cert_path = server_cert_path\n+ self.use_databricks_sdk = use_databricks_sdk\n+ self.databricks_auth_profile = databricks_auth_profile\n+ self.client_id = client_id\n+ self.client_secret = client_secret\n+ self.use_secret_scope_token = use_secret_scope_token\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,11 +712,12 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n+ @property\n+ def verify(self):\n+ if self.use_databricks_sdk:\n+ # Let databricks-sdk set HTTP request `verify` param.\n+ return None\n+ if self.server_cert_path is None:\n+ return not self.ignore_tls_verification\n+ else:\n+ return self.server_cert_path\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..dbb13d2e5 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -851,6 +851,9 @@ class ModelRegistryClient:\n Returns:\n A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n \"\"\"\n+ if max_results is None:\n+ # No result limit requested, so read the listing from the beginning.\n+ return self.store.list_webhooks(max_results, None)\n return self.store.list_webhooks(max_results, page_token)\n \n @experimental(version=\"3.3.0\")\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py\ndeleted file mode 100644\nindex 97fd61b00..000000000\n--- a/tests/tracking/test_client_webhooks.py\n+++ /dev/null\n@@ -1,219 +0,0 @@\n-from pathlib import Path\n-from typing import Iterator\n-\n-import pytest\n-from cryptography.fernet import Fernet\n-\n-from mlflow.entities.webhook import WebhookAction, WebhookEntity, WebhookEvent, WebhookStatus\n-from mlflow.environment_variables import MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY\n-from mlflow.exceptions import MlflowException\n-from mlflow.server import handlers\n-from mlflow.server.fastapi_app import app\n-from mlflow.server.handlers import initialize_backend_stores\n-from mlflow.tracking import MlflowClient\n-\n-from tests.helper_functions import get_safe_port\n-from tests.tracking.integration_test_utils import ServerThread\n-\n-\n-@pytest.fixture\n-def client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Iterator[MlflowClient]:\n- \"\"\"Setup a local MLflow server with proper webhook encryption key support.\"\"\"\n- # Set up encryption key for webhooks using monkeypatch\n- encryption_key = Fernet.generate_key().decode(\"utf-8\")\n- monkeypatch.setenv(MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY.name, encryption_key)\n-\n- # Configure backend stores\n- backend_uri = f\"sqlite:///{tmp_path / 'mlflow.db'}\"\n- default_artifact_root = tmp_path.as_uri()\n-\n- # Force-reset backend stores before each test\n- handlers._tracking_store = None\n- handlers._model_registry_store = None\n- initialize_backend_stores(backend_uri, default_artifact_root=default_artifact_root)\n-\n- # Start server and return client\n- with ServerThread(app, get_safe_port()) as url:\n- yield MlflowClient(url)\n-\n-\n-def test_create_webhook(client: MlflowClient):\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n- assert webhook.events == [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n-\n- webhook = client.get_webhook(webhook.webhook_id)\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n-\n- # With secret\n- webhook_with_secret = client.create_webhook(\n- name=\"test_webhook_with_secret\",\n- url=\"https://example.com/webhook_with_secret\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- secret=\"my_secret\",\n- )\n- assert webhook_with_secret.name == \"test_webhook_with_secret\"\n- assert webhook_with_secret.url == \"https://example.com/webhook_with_secret\"\n- assert webhook_with_secret.secret is None\n- assert webhook_with_secret.events == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)\n- ]\n-\n- # Multiple events\n- webhook_multiple_events = client.create_webhook(\n- name=\"test_webhook_multiple_events\",\n- url=\"https://example.com/webhook_multiple_events\",\n- events=[\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- ],\n- )\n- assert webhook_multiple_events.name == \"test_webhook_multiple_events\"\n- assert webhook_multiple_events.url == \"https://example.com/webhook_multiple_events\"\n- assert sorted(\n- webhook_multiple_events.events, key=lambda e: (e.entity.value, e.action.value)\n- ) == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- ]\n- assert webhook_multiple_events.secret is None\n-\n-\n-def test_get_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- created_webhook = client.create_webhook(\n- name=\"test_webhook\", url=\"https://example.com/webhook\", events=events\n- )\n- retrieved_webhook = client.get_webhook(created_webhook.webhook_id)\n- assert retrieved_webhook.webhook_id == created_webhook.webhook_id\n- assert retrieved_webhook.name == \"test_webhook\"\n- assert retrieved_webhook.url == \"https://example.com/webhook\"\n- assert retrieved_webhook.events == events\n-\n-\n-def test_get_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.get_webhook(\"nonexistent\")\n-\n-\n-def test_list_webhooks(client: MlflowClient):\n- # Create more webhooks than max_results\n- created_webhooks = []\n- for i in range(5):\n- webhook = client.create_webhook(\n- name=f\"webhook{i}\",\n- url=f\"https://example.com/{i}\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- created_webhooks.append(webhook)\n- # Test pagination with max_results=2\n- webhooks_page = client.list_webhooks(max_results=2)\n- assert len(webhooks_page) == 2\n- assert webhooks_page.token is not None\n- # Get next page\n- next_webhooks_page = client.list_webhooks(max_results=2, page_token=webhooks_page.token)\n- assert len(next_webhooks_page) == 2\n- assert next_webhooks_page.token is not None\n- # Verify we don't get duplicates\n- first_page_ids = {w.webhook_id for w in webhooks_page}\n- second_page_ids = {w.webhook_id for w in next_webhooks_page}\n- assert first_page_ids.isdisjoint(second_page_ids)\n-\n-\n-def test_update_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\", url=\"https://example.com/original\", events=events\n- )\n- # Update webhook\n- new_events = [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.REGISTERED_MODEL, WebhookAction.CREATED),\n- ]\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- url=\"https://example.com/updated\",\n- events=new_events,\n- description=\"Updated description\",\n- secret=\"new_secret\",\n- status=WebhookStatus.DISABLED,\n- )\n- assert updated_webhook.webhook_id == webhook.webhook_id\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/updated\"\n- assert updated_webhook.events == new_events\n- assert updated_webhook.description == \"Updated description\"\n- assert updated_webhook.status == WebhookStatus.DISABLED\n- assert updated_webhook.last_updated_timestamp > webhook.last_updated_timestamp\n-\n-\n-def test_update_webhook_partial(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\",\n- url=\"https://example.com/original\",\n- events=events,\n- description=\"Original description\",\n- )\n- # Update only the name\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- )\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/original\"\n- assert updated_webhook.events == events\n- assert updated_webhook.description == \"Original description\"\n-\n-\n-def test_update_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.update_webhook(webhook_id=\"nonexistent\", name=\"new_name\")\n-\n-\n-@pytest.mark.parametrize(\n- (\"invalid_url\", \"expected_match\"),\n- [\n- (\" \", r\"Webhook URL cannot be empty or just whitespace\"),\n- (\"ftp://example.com\", r\"Invalid webhook URL scheme\"),\n- (\"http://[invalid\", r\"Invalid webhook URL\"),\n- ],\n-)\n-def test_update_webhook_invalid_urls(client: MlflowClient, invalid_url: str, expected_match: str):\n- # Create a valid webhook first\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- with pytest.raises(MlflowException, match=expected_match):\n- client.update_webhook(webhook_id=webhook.webhook_id, url=invalid_url)\n-\n-\n-def test_delete_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=events,\n- )\n- client.delete_webhook(webhook.webhook_id)\n- with pytest.raises(MlflowException, match=r\"Webhook with ID .* not found\"):\n- client.get_webhook(webhook.webhook_id)\n- webhooks_page = client.list_webhooks()\n- webhook_ids = {w.webhook_id for w in webhooks_page}\n- assert webhook.webhook_id not in webhook_ids\n-\n-\n-def test_delete_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.delete_webhook(\"nonexistent\")\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S11\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":59,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S9\",\n \"text\": \"Only the fields you specify will be updated; other fields will remain unchanged. name: If None, the current name is preserved. description: If None, the current description is preserved. url: If None, the current URL is preserved. events: If None, the current events are preserved. secret: New secret for HMAC signature verification. If None, the current secret is preserved. Pass an empty string to remove the secret. status: If None, the current status is preserved. Task statement: Maintain webhook metadata consistency across creation, updates, and deletions.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracking/client.py\",\n \"lines\": \"6418-6425\",\n \"function\": \"MlflowClient.update_webhook\"\n },\n {\n \"file\": \"mlflow/tracking/_model_registry/client.py\",\n \"lines\": \"882\",\n \"function\": \"ModelRegistryClient.update_webhook\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n--- a/mlflow/entities/webhook.py\n+++ b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n- def __eq__(self, other: object) -> bool:\n- if not isinstance(other, WebhookEvent):\n- return False\n- return self.entity == other.entity and self.action == other.action\n+\n+\n+\n+\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n- @property\n- def secret(self) -> str | None:\n- return self._secret\n+\n+\n+\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n--- a/mlflow/server/handlers.py\n+++ b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n--- a/mlflow/store/artifact/azure_blob_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-from mlflow.utils.credentials import get_default_host_creds\n+\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- augmented_raise_for_status,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n- def __init__(self, store_uri=None, tracking_uri=None):\n- \"\"\"\n- Empty constructor. This is deliberately not marked as abstract, else every derived class\n- would be forced to create one.\n \n- Args:\n- store_uri: The model registry store URI.\n- tracking_uri: URI of the current MLflow tracking server, used to perform operations\n- like fetching source run metadata or downloading source run artifacts\n- to support subsequently uploading them to the model registry storage\n- location.\n- \"\"\"\n- # Create a thread lock to ensure thread safety when linking prompts to other entities,\n- # since the default linking implementation reads and appends entity tags, which\n- # is prone to concurrent modification issues\n- self._prompt_link_lock = threading.RLock()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n--- a/mlflow/store/model_registry/base_rest_store.py\n+++ b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n-from mlflow.utils.rest_utils import (\n- call_endpoint,\n- call_endpoints,\n-)\n+from mlflow.utils.rest_utils import call_endpoints\n+\n+\n+\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- @record_usage_event(CreateWebhookEvent)\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Unique name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of event types that trigger this webhook.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE).\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- return self.store.create_webhook(name, url, events, description, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self.store.get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self.store.list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of event types.\n- secret: New webhook secret.\n- status: New webhook status.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self.store.delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n- return self.store.test_webhook(webhook_id, event)\n+ return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n--- a/mlflow/tracking/_model_registry/registry.py\n+++ b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n- from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n+\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n--- a/mlflow/tracking/_model_registry/utils.py\n+++ b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n-def _get_registry_uri_from_spark_session():\n- session = _get_active_spark_session()\n- if session is None:\n- return None\n \n- if is_in_databricks_serverless_runtime():\n- # Connected to Serverless\n- return \"databricks-uc\"\n \n- from pyspark.sql.utils import AnalysisException\n \n- try:\n- return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n- except AnalysisException:\n- # In serverless clusters, session.conf.get() is unsupported\n- # and raises an AnalysisException. We may encounter this case\n- # when DBConnect is used to connect to a serverless cluster,\n- # in which case the prior `is_in_databricks_serverless_runtime()`\n- # check will have returned false (as of 2025-06-07, it checks\n- # an environment variable that isn't set by DBConnect)\n- return None\n \n \n-def _get_registry_uri_from_context():\n- if _registry_uri is not None:\n- return _registry_uri\n- elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n- return uri\n- return _registry_uri\n \n \n-def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n- \"\"\"\n- Get the default registry URI for a given tracking URI.\n \n- If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n- Otherwise, returns the tracking URI itself.\n \n- Args:\n- tracking_uri: The tracking URI to get the default registry URI for\n \n- Returns:\n- The default registry URI\n- \"\"\"\n- if tracking_uri is not None and is_databricks_uri(tracking_uri):\n- # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n- # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n- # model registry offering on Databricks\n- if tracking_uri == \"databricks\":\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n- else:\n- # Extract profile from tracking URI and construct databricks-uc URI\n- profile, key_prefix = get_db_info_from_uri(tracking_uri)\n- if profile:\n- # Reconstruct the profile string including key_prefix if present\n- profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n- return construct_db_uc_uri_from_profile(profile_string)\n- else:\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n-\n- # For non-databricks tracking URIs, use the tracking URI as the registry URI\n- return tracking_uri\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n-def _resolve_registry_uri(\n- registry_uri: str | None = None, tracking_uri: str | None = None\n-) -> str | None:\n- \"\"\"\n- Resolve the registry URI following the same logic as get_registry_uri().\n- \"\"\"\n- return (\n- registry_uri\n- or _get_registry_uri_from_context()\n- or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n-def _get_rest_store(store_uri, **_):\n- return RestStore(partial(get_default_host_creds, store_uri))\n+\n+\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n-def _get_store_registry():\n- global _model_registry_store_registry\n- from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n- from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n- if _model_registry_store_registry is not None:\n- return _model_registry_store_registry\n \n- _model_registry_store_registry = ModelRegistryStoreRegistry()\n- _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n- # Register a placeholder function that raises if users pass a registry URI with scheme\n- # \"databricks-uc\"\n- _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n- _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n- for scheme in [\"http\", \"https\"]:\n- _model_registry_store_registry.register(scheme, _get_rest_store)\n \n- for scheme in DATABASE_ENGINES:\n- _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n- for scheme in [\"\", \"file\"]:\n- _model_registry_store_registry.register(scheme, _get_file_store)\n \n- _model_registry_store_registry.register_entrypoints()\n- return _model_registry_store_registry\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n--- a/mlflow/tracking/_tracking_service/utils.py\n+++ b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n- @property\n- def tracking_uri(self):\n- return self._tracking_client.tracking_uri\n \n- def _get_registry_client(self):\n- \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n- Raises:\n- MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n- example, when the registry URI refers to an unsupported store type (e.g., the\n- FileStore).\n \n- Returns:\n- A ModelRegistryClient instance.\n- \"\"\"\n- # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n- # an instance variable on this `MlflowClient` instance. Because the instance variable\n- # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n- # function is used to safely fetch the variable (if it is defined) or a NoneType\n- # (if it is not defined)\n- registry_client_attr = \"_registry_client_lazy\"\n- registry_client = getattr(self, registry_client_attr, None)\n- if registry_client is None:\n- try:\n- registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n- # Define an instance variable on this `MlflowClient` instance to reference the\n- # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n- # that the variable name is consistent with the variable name specified in the\n- # preceding call to `getattr()`\n- setattr(self, registry_client_attr, registry_client)\n- except UnsupportedModelRegistryStoreURIException as exc:\n- raise MlflowException(\n- \"Model Registry features are not supported by the store with URI:\"\n- f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n- f\" {exc.supported_uri_schemes}.\",\n- FEATURE_DISABLED,\n- )\n- return registry_client\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEventStr | WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of events that trigger this webhook. Can be strings or\n- `WebhookEvent` objects.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self._get_registry_client().get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self._get_registry_client().list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEventStr | WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of events. Can be strings or `WebhookEvent` objects.\n- secret: New webhook secret.\n- status: New webhook status. Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().update_webhook(\n- webhook_id=webhook_id,\n- name=name,\n- description=description,\n- url=url,\n- events=events,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self._get_registry_client().delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n- return self._get_registry_client().test_webhook(webhook_id, event)\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,7 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,36 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n--- a/mlflow/utils/credentials.py\n+++ b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n-def get_default_host_creds(store_uri):\n- creds = read_mlflow_creds()\n- return MlflowHostCreds(\n- host=store_uri,\n- username=creds.username,\n- password=creds.password,\n- token=MLFLOW_TRACKING_TOKEN.get(),\n- aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n- auth=MLFLOW_TRACKING_AUTH.get(),\n- ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n- client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n- server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n-def parse_dict(js_dict, message):\n- \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n- _stringify_all_experiment_ids(js_dict)\n- ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n+\n+\n+\n+\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n--- a/mlflow/utils/request_utils.py\n+++ b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n-@lru_cache(maxsize=64)\n-def _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- # To create a new Session object for each process, we use the process id as the cache key.\n- # This is to avoid sharing the same Session object across processes, which can lead to issues\n- # such as https://stackoverflow.com/q/3724900.\n- _pid,\n- respect_retry_after_header=True,\n-):\n- \"\"\"\n- This function should not be called directly. Instead, use `_get_request_session` below.\n- \"\"\"\n \n- retry_kwargs = {\n- \"total\": max_retries,\n- \"connect\": max_retries,\n- \"read\": max_retries,\n- \"redirect\": max_retries,\n- \"status\": max_retries,\n- \"status_forcelist\": retry_codes,\n- \"backoff_factor\": backoff_factor,\n- \"backoff_jitter\": backoff_jitter,\n- \"raise_on_status\": raise_on_status,\n- \"respect_retry_after_header\": respect_retry_after_header,\n- }\n- urllib3_version = Version(urllib3.__version__)\n- if urllib3_version >= Version(\"1.26.0\"):\n- retry_kwargs[\"allowed_methods\"] = None\n- else:\n- retry_kwargs[\"method_whitelist\"] = None\n-\n- if urllib3_version < Version(\"2.0\"):\n- retry = JitteredRetry(**retry_kwargs)\n- else:\n- retry = Retry(**retry_kwargs)\n- from mlflow.environment_variables import (\n- MLFLOW_HTTP_POOL_CONNECTIONS,\n- MLFLOW_HTTP_POOL_MAXSIZE,\n- )\n-\n- adapter = HTTPAdapter(\n- pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n- pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n- max_retries=retry,\n- )\n- session = requests.Session()\n- session.mount(\"https://\", adapter)\n- session.mount(\"http://\", adapter)\n- return session\n-\n-\n-def _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n-):\n- \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n- Args:\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n- Returns:\n- requests.Session object.\n \n- \"\"\"\n- return _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- _pid=os.getpid(),\n- respect_retry_after_header=respect_retry_after_header,\n- )\n \n \n-def _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status=True,\n- allow_redirects=None,\n- respect_retry_after_header=True,\n- **kwargs,\n-):\n- \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n- Args:\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- url: The target URL address for the HTTP request.\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- session = _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n- )\n \n- # the environment variable is hardcoded here to avoid importing mlflow.\n- # however, documentation is available in environment_variables.py\n- env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n- allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n- return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n-from mlflow.utils.proto_json_utils import parse_dict\n-from mlflow.utils.request_utils import (\n- _TRANSIENT_FAILURE_RESPONSE_CODES,\n- _get_http_response_with_retries,\n- augmented_raise_for_status, # noqa: F401\n- cloud_storage_http_request, # noqa: F401\n-)\n-from mlflow.utils.string_utils import strip_suffix\n+\n+from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n+\n+\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n \n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n \n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n-\n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n \n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n-\n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n-\n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n-\n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n-\n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n-\n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n-\n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n-def _can_parse_as_json_object(string):\n- try:\n- return isinstance(json.loads(string), dict)\n- except Exception:\n- return False\n+\n+\n+\n+\n+\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n-def verify_rest_response(response, endpoint):\n- \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n- # Handle Armeria-specific response case where response text is \"200 OK\"\n- # v1/traces endpoint might return empty response\n- if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n- response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n- return response\n-\n- # Handle non-200 status codes\n- if response.status_code != 200:\n- if _can_parse_as_json_object(response.text):\n- raise RestException(json.loads(response.text))\n- else:\n- base_msg = (\n- f\"API request to endpoint {endpoint} \"\n- f\"failed with error code {response.status_code} != 200\"\n- )\n- raise MlflowException(\n- f\"{base_msg}. Response body: '{response.text}'\",\n- error_code=get_error_code(response.status_code),\n- )\n \n- # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n- # response\n- if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n- base_msg = (\n- \"API request to endpoint was successful but the response body was not \"\n- \"in a valid JSON format\"\n- )\n- raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n-\n- return response\n-\n-\n-def _validate_max_retries(max_retries):\n- max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n-\n- if max_retry_limit < 0:\n- raise MlflowException(\n- message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n- \"Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if max_retries >= max_retry_limit:\n- raise MlflowException(\n- message=f\"The configured max_retries value ({max_retries}) is \"\n- f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if max_retries < 0:\n- raise MlflowException(\n- message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n-\n-def _validate_backoff_factor(backoff_factor):\n- max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n-\n- if max_backoff_factor_limit < 0:\n- raise MlflowException(\n- message=\"The current maximum backoff factor limit is invalid \"\n- f\"({max_backoff_factor_limit}). Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor >= max_backoff_factor_limit:\n- raise MlflowException(\n- message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n- \"of the maximum allowable backoff_factor limit \"\n- f\"({max_backoff_factor_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor < 0:\n- raise MlflowException(\n- message=\"The backoff_factor value must be either 0 a positive integer. \"\n- f\"Got {backoff_factor}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n-def call_endpoint(\n- host_creds,\n- endpoint,\n- method,\n- json_body,\n- response_proto,\n- extra_headers=None,\n- retry_timeout_seconds=None,\n-):\n- # Convert json string to json dictionary, to pass to requests\n- if json_body is not None:\n- json_body = json.loads(json_body)\n- call_kwargs = {\n- \"host_creds\": host_creds,\n- \"endpoint\": endpoint,\n- \"method\": method,\n- }\n- if extra_headers is not None:\n- call_kwargs[\"extra_headers\"] = extra_headers\n- if retry_timeout_seconds is not None:\n- call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n- if method == \"GET\":\n- call_kwargs[\"params\"] = json_body\n- response = http_request(**call_kwargs)\n- else:\n- call_kwargs[\"json\"] = json_body\n- response = http_request(**call_kwargs)\n \n- response = verify_rest_response(response, endpoint)\n- response_to_parse = response.text\n- try:\n- js_dict = json.loads(response_to_parse)\n- except json.JSONDecodeError:\n- if len(response_to_parse) > 50:\n- response_to_parse = response_to_parse[:50] + \"...\"\n- _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n- raise\n \n- parse_dict(js_dict=js_dict, message=response_proto)\n- return response_proto\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n- def __init__(\n- self,\n- host,\n- username=None,\n- password=None,\n- token=None,\n- aws_sigv4=False,\n- auth=None,\n- ignore_tls_verification=False,\n- client_cert_path=None,\n- server_cert_path=None,\n- use_databricks_sdk=False,\n- databricks_auth_profile=None,\n- client_id=None,\n- client_secret=None,\n- use_secret_scope_token=False,\n- ):\n- if not host:\n- raise MlflowException(\n- message=\"host is a required parameter for MlflowHostCreds\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if ignore_tls_verification and (server_cert_path is not None):\n- raise MlflowException(\n- message=(\n- \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n- \"must not be set! This error may have occurred because the \"\n- \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n- \"environment variables are both set - only one of these environment \"\n- \"variables may be set.\"\n- ),\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- self.host = host\n- self.username = username\n- self.password = password\n- self.token = token\n- self.aws_sigv4 = aws_sigv4\n- self.auth = auth\n- self.ignore_tls_verification = ignore_tls_verification\n- self.client_cert_path = client_cert_path\n- self.server_cert_path = server_cert_path\n- self.use_databricks_sdk = use_databricks_sdk\n- self.databricks_auth_profile = databricks_auth_profile\n- self.client_id = client_id\n- self.client_secret = client_secret\n- self.use_secret_scope_token = use_secret_scope_token\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,12 +712,11 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n- @property\n- def verify(self):\n- if self.use_databricks_sdk:\n- # Let databricks-sdk set HTTP request `verify` param.\n- return None\n- if self.server_cert_path is None:\n- return not self.ignore_tls_verification\n- else:\n- return self.server_cert_path\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n+++ a/mlflow/entities/webhook.py\n--- b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n+ def __eq__(self, other: object) -> bool:\n+ if not isinstance(other, WebhookEvent):\n+ return False\n+ return self.entity == other.entity and self.action == other.action\n-\n-\n-\n-\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n+ @property\n+ def secret(self) -> str | None:\n+ return self._secret\n-\n-\n-\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n+++ a/mlflow/server/handlers.py\n--- b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n+++ a/mlflow/store/artifact/azure_blob_artifact_repo.py\n--- b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+from mlflow.utils.credentials import get_default_host_creds\n-\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ augmented_raise_for_status,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(self, store_uri=None, tracking_uri=None):\n+ \"\"\"\n+ Empty constructor. This is deliberately not marked as abstract, else every derived class\n+ would be forced to create one.\n \n+ Args:\n+ store_uri: The model registry store URI.\n+ tracking_uri: URI of the current MLflow tracking server, used to perform operations\n+ like fetching source run metadata or downloading source run artifacts\n+ to support subsequently uploading them to the model registry storage\n+ location.\n+ \"\"\"\n+ # Create a thread lock to ensure thread safety when linking prompts to other entities,\n+ # since the default linking implementation reads and appends entity tags, which\n+ # is prone to concurrent modification issues\n+ self._prompt_link_lock = threading.RLock()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n+++ a/mlflow/store/model_registry/base_rest_store.py\n--- b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n+from mlflow.utils.rest_utils import (\n+ call_endpoint,\n+ call_endpoints,\n+)\n-from mlflow.utils.rest_utils import call_endpoints\n-\n-\n-\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n+++ a/mlflow/tracking/_model_registry/client.py\n--- b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ @record_usage_event(CreateWebhookEvent)\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Unique name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of event types that trigger this webhook.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE).\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ return self.store.create_webhook(name, url, events, description, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self.store.get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self.store.list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of event types.\n+ secret: New webhook secret.\n+ status: New webhook status.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self.store.delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n+ return self.store.test_webhook(webhook_id, event)\n- return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n+++ a/mlflow/tracking/_model_registry/registry.py\n--- b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n+ from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n-\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n+++ a/mlflow/tracking/_model_registry/utils.py\n--- b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n+def _get_registry_uri_from_spark_session():\n+ session = _get_active_spark_session()\n+ if session is None:\n+ return None\n \n+ if is_in_databricks_serverless_runtime():\n+ # Connected to Serverless\n+ return \"databricks-uc\"\n \n+ from pyspark.sql.utils import AnalysisException\n \n+ try:\n+ return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n+ except AnalysisException:\n+ # In serverless clusters, session.conf.get() is unsupported\n+ # and raises an AnalysisException. We may encounter this case\n+ # when DBConnect is used to connect to a serverless cluster,\n+ # in which case the prior `is_in_databricks_serverless_runtime()`\n+ # check will have returned false (as of 2025-06-07, it checks\n+ # an environment variable that isn't set by DBConnect)\n+ return None\n \n \n+def _get_registry_uri_from_context():\n+ if _registry_uri is not None:\n+ return _registry_uri\n+ elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n+ return uri\n+ return _registry_uri\n \n \n+def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n+ \"\"\"\n+ Get the default registry URI for a given tracking URI.\n \n+ If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n+ Otherwise, returns the tracking URI itself.\n \n+ Args:\n+ tracking_uri: The tracking URI to get the default registry URI for\n \n+ Returns:\n+ The default registry URI\n+ \"\"\"\n+ if tracking_uri is not None and is_databricks_uri(tracking_uri):\n+ # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n+ # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n+ # model registry offering on Databricks\n+ if tracking_uri == \"databricks\":\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+ else:\n+ # Extract profile from tracking URI and construct databricks-uc URI\n+ profile, key_prefix = get_db_info_from_uri(tracking_uri)\n+ if profile:\n+ # Reconstruct the profile string including key_prefix if present\n+ profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n+ return construct_db_uc_uri_from_profile(profile_string)\n+ else:\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+\n+ # For non-databricks tracking URIs, use the tracking URI as the registry URI\n+ return tracking_uri\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n+def _resolve_registry_uri(\n+ registry_uri: str | None = None, tracking_uri: str | None = None\n+) -> str | None:\n+ \"\"\"\n+ Resolve the registry URI following the same logic as get_registry_uri().\n+ \"\"\"\n+ return (\n+ registry_uri\n+ or _get_registry_uri_from_context()\n+ or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n+def _get_rest_store(store_uri, **_):\n+ return RestStore(partial(get_default_host_creds, store_uri))\n-\n-\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n+def _get_store_registry():\n+ global _model_registry_store_registry\n+ from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n+ from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n+ if _model_registry_store_registry is not None:\n+ return _model_registry_store_registry\n \n+ _model_registry_store_registry = ModelRegistryStoreRegistry()\n+ _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n+ # Register a placeholder function that raises if users pass a registry URI with scheme\n+ # \"databricks-uc\"\n+ _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n+ _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n+ for scheme in [\"http\", \"https\"]:\n+ _model_registry_store_registry.register(scheme, _get_rest_store)\n \n+ for scheme in DATABASE_ENGINES:\n+ _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n+ for scheme in [\"\", \"file\"]:\n+ _model_registry_store_registry.register(scheme, _get_file_store)\n \n+ _model_registry_store_registry.register_entrypoints()\n+ return _model_registry_store_registry\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n+++ a/mlflow/tracking/_tracking_service/utils.py\n--- b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n+++ a/mlflow/tracking/client.py\n--- b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n+ @property\n+ def tracking_uri(self):\n+ return self._tracking_client.tracking_uri\n \n+ def _get_registry_client(self):\n+ \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n+ Raises:\n+ MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n+ example, when the registry URI refers to an unsupported store type (e.g., the\n+ FileStore).\n \n+ Returns:\n+ A ModelRegistryClient instance.\n+ \"\"\"\n+ # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n+ # an instance variable on this `MlflowClient` instance. Because the instance variable\n+ # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n+ # function is used to safely fetch the variable (if it is defined) or a NoneType\n+ # (if it is not defined)\n+ registry_client_attr = \"_registry_client_lazy\"\n+ registry_client = getattr(self, registry_client_attr, None)\n+ if registry_client is None:\n+ try:\n+ registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n+ # Define an instance variable on this `MlflowClient` instance to reference the\n+ # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n+ # that the variable name is consistent with the variable name specified in the\n+ # preceding call to `getattr()`\n+ setattr(self, registry_client_attr, registry_client)\n+ except UnsupportedModelRegistryStoreURIException as exc:\n+ raise MlflowException(\n+ \"Model Registry features are not supported by the store with URI:\"\n+ f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n+ f\" {exc.supported_uri_schemes}.\",\n+ FEATURE_DISABLED,\n+ )\n+ return registry_client\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEventStr | WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of events that trigger this webhook. Can be strings or\n+ `WebhookEvent` objects.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self._get_registry_client().get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self._get_registry_client().list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEventStr | WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of events. Can be strings or `WebhookEvent` objects.\n+ secret: New webhook secret.\n+ status: New webhook status. Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ if events is not None:\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().update_webhook(\n+ webhook_id=webhook_id,\n+ name=name,\n+ description=description,\n+ url=url,\n+ events=events,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self._get_registry_client().delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n- return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,7 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,36 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n+++ a/mlflow/utils/credentials.py\n--- b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n+def get_default_host_creds(store_uri):\n+ creds = read_mlflow_creds()\n+ return MlflowHostCreds(\n+ host=store_uri,\n+ username=creds.username,\n+ password=creds.password,\n+ token=MLFLOW_TRACKING_TOKEN.get(),\n+ aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n+ auth=MLFLOW_TRACKING_AUTH.get(),\n+ ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n+ client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n+ server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n+def parse_dict(js_dict, message):\n+ \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n+ _stringify_all_experiment_ids(js_dict)\n+ ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n-\n-\n-\n-\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n+++ a/mlflow/utils/request_utils.py\n--- b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n+@lru_cache(maxsize=64)\n+def _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ # To create a new Session object for each process, we use the process id as the cache key.\n+ # This is to avoid sharing the same Session object across processes, which can lead to issues\n+ # such as https://stackoverflow.com/q/3724900.\n+ _pid,\n+ respect_retry_after_header=True,\n+):\n+ \"\"\"\n+ This function should not be called directly. Instead, use `_get_request_session` below.\n+ \"\"\"\n \n+ retry_kwargs = {\n+ \"total\": max_retries,\n+ \"connect\": max_retries,\n+ \"read\": max_retries,\n+ \"redirect\": max_retries,\n+ \"status\": max_retries,\n+ \"status_forcelist\": retry_codes,\n+ \"backoff_factor\": backoff_factor,\n+ \"backoff_jitter\": backoff_jitter,\n+ \"raise_on_status\": raise_on_status,\n+ \"respect_retry_after_header\": respect_retry_after_header,\n+ }\n+ urllib3_version = Version(urllib3.__version__)\n+ if urllib3_version >= Version(\"1.26.0\"):\n+ retry_kwargs[\"allowed_methods\"] = None\n+ else:\n+ retry_kwargs[\"method_whitelist\"] = None\n+\n+ if urllib3_version < Version(\"2.0\"):\n+ retry = JitteredRetry(**retry_kwargs)\n+ else:\n+ retry = Retry(**retry_kwargs)\n+ from mlflow.environment_variables import (\n+ MLFLOW_HTTP_POOL_CONNECTIONS,\n+ MLFLOW_HTTP_POOL_MAXSIZE,\n+ )\n+\n+ adapter = HTTPAdapter(\n+ pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n+ pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n+ max_retries=retry,\n+ )\n+ session = requests.Session()\n+ session.mount(\"https://\", adapter)\n+ session.mount(\"http://\", adapter)\n+ return session\n+\n+\n+def _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+):\n+ \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n+ Args:\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n+ Returns:\n+ requests.Session object.\n \n+ \"\"\"\n+ return _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ _pid=os.getpid(),\n+ respect_retry_after_header=respect_retry_after_header,\n+ )\n \n \n+def _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status=True,\n+ allow_redirects=None,\n+ respect_retry_after_header=True,\n+ **kwargs,\n+):\n+ \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n+ Args:\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ url: The target URL address for the HTTP request.\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ session = _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+ )\n \n+ # the environment variable is hardcoded here to avoid importing mlflow.\n+ # however, documentation is available in environment_variables.py\n+ env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n+ allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n+ return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n+from mlflow.utils.proto_json_utils import parse_dict\n+from mlflow.utils.request_utils import (\n+ _TRANSIENT_FAILURE_RESPONSE_CODES,\n+ _get_http_response_with_retries,\n+ augmented_raise_for_status, # noqa: F401\n+ cloud_storage_http_request, # noqa: F401\n+)\n+from mlflow.utils.string_utils import strip_suffix\n-\n-from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n-\n-\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n \n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n \n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n+\n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n \n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n+\n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n+\n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n+\n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n+\n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n+\n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n+\n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n+def _can_parse_as_json_object(string):\n+ try:\n+ return isinstance(json.loads(string), dict)\n+ except Exception:\n+ return False\n-\n-\n-\n-\n-\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n+def verify_rest_response(response, endpoint):\n+ \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n+ # Handle Armeria-specific response case where response text is \"200 OK\"\n+ # v1/traces endpoint might return empty response\n+ if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n+ response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n+ return response\n+\n+ # Handle non-200 status codes\n+ if response.status_code != 200:\n+ if _can_parse_as_json_object(response.text):\n+ raise RestException(json.loads(response.text))\n+ else:\n+ base_msg = (\n+ f\"API request to endpoint {endpoint} \"\n+ f\"failed with error code {response.status_code} != 200\"\n+ )\n+ raise MlflowException(\n+ f\"{base_msg}. Response body: '{response.text}'\",\n+ error_code=get_error_code(response.status_code),\n+ )\n \n+ # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n+ # response\n+ if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n+ base_msg = (\n+ \"API request to endpoint was successful but the response body was not \"\n+ \"in a valid JSON format\"\n+ )\n+ raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n+\n+ return response\n+\n+\n+def _validate_max_retries(max_retries):\n+ max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n+\n+ if max_retry_limit < 0:\n+ raise MlflowException(\n+ message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n+ \"Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if max_retries >= max_retry_limit:\n+ raise MlflowException(\n+ message=f\"The configured max_retries value ({max_retries}) is \"\n+ f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if max_retries < 0:\n+ raise MlflowException(\n+ message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+\n+def _validate_backoff_factor(backoff_factor):\n+ max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n+\n+ if max_backoff_factor_limit < 0:\n+ raise MlflowException(\n+ message=\"The current maximum backoff factor limit is invalid \"\n+ f\"({max_backoff_factor_limit}). Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor >= max_backoff_factor_limit:\n+ raise MlflowException(\n+ message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n+ \"of the maximum allowable backoff_factor limit \"\n+ f\"({max_backoff_factor_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor < 0:\n+ raise MlflowException(\n+ message=\"The backoff_factor value must be either 0 a positive integer. \"\n+ f\"Got {backoff_factor}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n+def call_endpoint(\n+ host_creds,\n+ endpoint,\n+ method,\n+ json_body,\n+ response_proto,\n+ extra_headers=None,\n+ retry_timeout_seconds=None,\n+):\n+ # Convert json string to json dictionary, to pass to requests\n+ if json_body is not None:\n+ json_body = json.loads(json_body)\n+ call_kwargs = {\n+ \"host_creds\": host_creds,\n+ \"endpoint\": endpoint,\n+ \"method\": method,\n+ }\n+ if extra_headers is not None:\n+ call_kwargs[\"extra_headers\"] = extra_headers\n+ if retry_timeout_seconds is not None:\n+ call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n+ if method == \"GET\":\n+ call_kwargs[\"params\"] = json_body\n+ response = http_request(**call_kwargs)\n+ else:\n+ call_kwargs[\"json\"] = json_body\n+ response = http_request(**call_kwargs)\n \n+ response = verify_rest_response(response, endpoint)\n+ response_to_parse = response.text\n+ try:\n+ js_dict = json.loads(response_to_parse)\n+ except json.JSONDecodeError:\n+ if len(response_to_parse) > 50:\n+ response_to_parse = response_to_parse[:50] + \"...\"\n+ _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n+ raise\n \n+ parse_dict(js_dict=js_dict, message=response_proto)\n+ return response_proto\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n+ def __init__(\n+ self,\n+ host,\n+ username=None,\n+ password=None,\n+ token=None,\n+ aws_sigv4=False,\n+ auth=None,\n+ ignore_tls_verification=False,\n+ client_cert_path=None,\n+ server_cert_path=None,\n+ use_databricks_sdk=False,\n+ databricks_auth_profile=None,\n+ client_id=None,\n+ client_secret=None,\n+ use_secret_scope_token=False,\n+ ):\n+ if not host:\n+ raise MlflowException(\n+ message=\"host is a required parameter for MlflowHostCreds\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if ignore_tls_verification and (server_cert_path is not None):\n+ raise MlflowException(\n+ message=(\n+ \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n+ \"must not be set! This error may have occurred because the \"\n+ \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n+ \"environment variables are both set - only one of these environment \"\n+ \"variables may be set.\"\n+ ),\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ self.host = host\n+ self.username = username\n+ self.password = password\n+ self.token = token\n+ self.aws_sigv4 = aws_sigv4\n+ self.auth = auth\n+ self.ignore_tls_verification = ignore_tls_verification\n+ self.client_cert_path = client_cert_path\n+ self.server_cert_path = server_cert_path\n+ self.use_databricks_sdk = use_databricks_sdk\n+ self.databricks_auth_profile = databricks_auth_profile\n+ self.client_id = client_id\n+ self.client_secret = client_secret\n+ self.use_secret_scope_token = use_secret_scope_token\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,11 +712,12 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n+ @property\n+ def verify(self):\n+ if self.use_databricks_sdk:\n+ # Let databricks-sdk set HTTP request `verify` param.\n+ return None\n+ if self.server_cert_path is None:\n+ return not self.ignore_tls_verification\n+ else:\n+ return self.server_cert_path\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cbb34f47f 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -6409,6 +6409,8 @@ class MlflowClient:\n \n if status is not None:\n status = WebhookStatus(status) if isinstance(status, str) else status\n+ else:\n+ status = WebhookStatus.ACTIVE\n \n return self._get_registry_client().update_webhook(\n webhook_id=webhook_id,\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py\ndeleted file mode 100644\nindex 97fd61b00..000000000\n--- a/tests/tracking/test_client_webhooks.py\n+++ /dev/null\n@@ -1,219 +0,0 @@\n-from pathlib import Path\n-from typing import Iterator\n-\n-import pytest\n-from cryptography.fernet import Fernet\n-\n-from mlflow.entities.webhook import WebhookAction, WebhookEntity, WebhookEvent, WebhookStatus\n-from mlflow.environment_variables import MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY\n-from mlflow.exceptions import MlflowException\n-from mlflow.server import handlers\n-from mlflow.server.fastapi_app import app\n-from mlflow.server.handlers import initialize_backend_stores\n-from mlflow.tracking import MlflowClient\n-\n-from tests.helper_functions import get_safe_port\n-from tests.tracking.integration_test_utils import ServerThread\n-\n-\n-@pytest.fixture\n-def client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Iterator[MlflowClient]:\n- \"\"\"Setup a local MLflow server with proper webhook encryption key support.\"\"\"\n- # Set up encryption key for webhooks using monkeypatch\n- encryption_key = Fernet.generate_key().decode(\"utf-8\")\n- monkeypatch.setenv(MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY.name, encryption_key)\n-\n- # Configure backend stores\n- backend_uri = f\"sqlite:///{tmp_path / 'mlflow.db'}\"\n- default_artifact_root = tmp_path.as_uri()\n-\n- # Force-reset backend stores before each test\n- handlers._tracking_store = None\n- handlers._model_registry_store = None\n- initialize_backend_stores(backend_uri, default_artifact_root=default_artifact_root)\n-\n- # Start server and return client\n- with ServerThread(app, get_safe_port()) as url:\n- yield MlflowClient(url)\n-\n-\n-def test_create_webhook(client: MlflowClient):\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n- assert webhook.events == [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n-\n- webhook = client.get_webhook(webhook.webhook_id)\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n-\n- # With secret\n- webhook_with_secret = client.create_webhook(\n- name=\"test_webhook_with_secret\",\n- url=\"https://example.com/webhook_with_secret\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- secret=\"my_secret\",\n- )\n- assert webhook_with_secret.name == \"test_webhook_with_secret\"\n- assert webhook_with_secret.url == \"https://example.com/webhook_with_secret\"\n- assert webhook_with_secret.secret is None\n- assert webhook_with_secret.events == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)\n- ]\n-\n- # Multiple events\n- webhook_multiple_events = client.create_webhook(\n- name=\"test_webhook_multiple_events\",\n- url=\"https://example.com/webhook_multiple_events\",\n- events=[\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- ],\n- )\n- assert webhook_multiple_events.name == \"test_webhook_multiple_events\"\n- assert webhook_multiple_events.url == \"https://example.com/webhook_multiple_events\"\n- assert sorted(\n- webhook_multiple_events.events, key=lambda e: (e.entity.value, e.action.value)\n- ) == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- ]\n- assert webhook_multiple_events.secret is None\n-\n-\n-def test_get_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- created_webhook = client.create_webhook(\n- name=\"test_webhook\", url=\"https://example.com/webhook\", events=events\n- )\n- retrieved_webhook = client.get_webhook(created_webhook.webhook_id)\n- assert retrieved_webhook.webhook_id == created_webhook.webhook_id\n- assert retrieved_webhook.name == \"test_webhook\"\n- assert retrieved_webhook.url == \"https://example.com/webhook\"\n- assert retrieved_webhook.events == events\n-\n-\n-def test_get_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.get_webhook(\"nonexistent\")\n-\n-\n-def test_list_webhooks(client: MlflowClient):\n- # Create more webhooks than max_results\n- created_webhooks = []\n- for i in range(5):\n- webhook = client.create_webhook(\n- name=f\"webhook{i}\",\n- url=f\"https://example.com/{i}\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- created_webhooks.append(webhook)\n- # Test pagination with max_results=2\n- webhooks_page = client.list_webhooks(max_results=2)\n- assert len(webhooks_page) == 2\n- assert webhooks_page.token is not None\n- # Get next page\n- next_webhooks_page = client.list_webhooks(max_results=2, page_token=webhooks_page.token)\n- assert len(next_webhooks_page) == 2\n- assert next_webhooks_page.token is not None\n- # Verify we don't get duplicates\n- first_page_ids = {w.webhook_id for w in webhooks_page}\n- second_page_ids = {w.webhook_id for w in next_webhooks_page}\n- assert first_page_ids.isdisjoint(second_page_ids)\n-\n-\n-def test_update_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\", url=\"https://example.com/original\", events=events\n- )\n- # Update webhook\n- new_events = [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.REGISTERED_MODEL, WebhookAction.CREATED),\n- ]\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- url=\"https://example.com/updated\",\n- events=new_events,\n- description=\"Updated description\",\n- secret=\"new_secret\",\n- status=WebhookStatus.DISABLED,\n- )\n- assert updated_webhook.webhook_id == webhook.webhook_id\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/updated\"\n- assert updated_webhook.events == new_events\n- assert updated_webhook.description == \"Updated description\"\n- assert updated_webhook.status == WebhookStatus.DISABLED\n- assert updated_webhook.last_updated_timestamp > webhook.last_updated_timestamp\n-\n-\n-def test_update_webhook_partial(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\",\n- url=\"https://example.com/original\",\n- events=events,\n- description=\"Original description\",\n- )\n- # Update only the name\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- )\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/original\"\n- assert updated_webhook.events == events\n- assert updated_webhook.description == \"Original description\"\n-\n-\n-def test_update_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.update_webhook(webhook_id=\"nonexistent\", name=\"new_name\")\n-\n-\n-@pytest.mark.parametrize(\n- (\"invalid_url\", \"expected_match\"),\n- [\n- (\" \", r\"Webhook URL cannot be empty or just whitespace\"),\n- (\"ftp://example.com\", r\"Invalid webhook URL scheme\"),\n- (\"http://[invalid\", r\"Invalid webhook URL\"),\n- ],\n-)\n-def test_update_webhook_invalid_urls(client: MlflowClient, invalid_url: str, expected_match: str):\n- # Create a valid webhook first\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- with pytest.raises(MlflowException, match=expected_match):\n- client.update_webhook(webhook_id=webhook.webhook_id, url=invalid_url)\n-\n-\n-def test_delete_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=events,\n- )\n- client.delete_webhook(webhook.webhook_id)\n- with pytest.raises(MlflowException, match=r\"Webhook with ID .* not found\"):\n- client.get_webhook(webhook.webhook_id)\n- webhooks_page = client.list_webhooks()\n- webhook_ids = {w.webhook_id for w in webhooks_page}\n- assert webhook.webhook_id not in webhook_ids\n-\n-\n-def test_delete_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.delete_webhook(\"nonexistent\")\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S8\",\n \"shared_functions\": 2,\n \"shared_lines\": 9\n },\n {\n \"spec_id\": \"S12\",\n \"shared_functions\": 2,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S11\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n }\n ]\n}"}}},{"rowIdx":60,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S12\",\n \"text\": \"Raises: mlflow.MlflowException: If the webhook with the specified ID does not exist, if the provided parameters are invalid, or if there are permission issues accessing the webhook.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracking/client.py\",\n \"lines\": \"6420-6425\",\n \"function\": \"MlflowClient.update_webhook\"\n },\n {\n \"file\": \"mlflow/tracking/_model_registry/client.py\",\n \"lines\": \"882\",\n \"function\": \"ModelRegistryClient.update_webhook\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n--- a/mlflow/entities/webhook.py\n+++ b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n- def __eq__(self, other: object) -> bool:\n- if not isinstance(other, WebhookEvent):\n- return False\n- return self.entity == other.entity and self.action == other.action\n+\n+\n+\n+\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n- @property\n- def secret(self) -> str | None:\n- return self._secret\n+\n+\n+\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n--- a/mlflow/server/handlers.py\n+++ b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n--- a/mlflow/store/artifact/azure_blob_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-from mlflow.utils.credentials import get_default_host_creds\n+\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- augmented_raise_for_status,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n- def __init__(self, store_uri=None, tracking_uri=None):\n- \"\"\"\n- Empty constructor. This is deliberately not marked as abstract, else every derived class\n- would be forced to create one.\n \n- Args:\n- store_uri: The model registry store URI.\n- tracking_uri: URI of the current MLflow tracking server, used to perform operations\n- like fetching source run metadata or downloading source run artifacts\n- to support subsequently uploading them to the model registry storage\n- location.\n- \"\"\"\n- # Create a thread lock to ensure thread safety when linking prompts to other entities,\n- # since the default linking implementation reads and appends entity tags, which\n- # is prone to concurrent modification issues\n- self._prompt_link_lock = threading.RLock()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n--- a/mlflow/store/model_registry/base_rest_store.py\n+++ b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n-from mlflow.utils.rest_utils import (\n- call_endpoint,\n- call_endpoints,\n-)\n+from mlflow.utils.rest_utils import call_endpoints\n+\n+\n+\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- @record_usage_event(CreateWebhookEvent)\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Unique name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of event types that trigger this webhook.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE).\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- return self.store.create_webhook(name, url, events, description, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self.store.get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self.store.list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of event types.\n- secret: New webhook secret.\n- status: New webhook status.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self.store.delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n- return self.store.test_webhook(webhook_id, event)\n+ return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n--- a/mlflow/tracking/_model_registry/registry.py\n+++ b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n- from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n+\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n--- a/mlflow/tracking/_model_registry/utils.py\n+++ b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n-def _get_registry_uri_from_spark_session():\n- session = _get_active_spark_session()\n- if session is None:\n- return None\n \n- if is_in_databricks_serverless_runtime():\n- # Connected to Serverless\n- return \"databricks-uc\"\n \n- from pyspark.sql.utils import AnalysisException\n \n- try:\n- return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n- except AnalysisException:\n- # In serverless clusters, session.conf.get() is unsupported\n- # and raises an AnalysisException. We may encounter this case\n- # when DBConnect is used to connect to a serverless cluster,\n- # in which case the prior `is_in_databricks_serverless_runtime()`\n- # check will have returned false (as of 2025-06-07, it checks\n- # an environment variable that isn't set by DBConnect)\n- return None\n \n \n-def _get_registry_uri_from_context():\n- if _registry_uri is not None:\n- return _registry_uri\n- elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n- return uri\n- return _registry_uri\n \n \n-def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n- \"\"\"\n- Get the default registry URI for a given tracking URI.\n \n- If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n- Otherwise, returns the tracking URI itself.\n \n- Args:\n- tracking_uri: The tracking URI to get the default registry URI for\n \n- Returns:\n- The default registry URI\n- \"\"\"\n- if tracking_uri is not None and is_databricks_uri(tracking_uri):\n- # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n- # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n- # model registry offering on Databricks\n- if tracking_uri == \"databricks\":\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n- else:\n- # Extract profile from tracking URI and construct databricks-uc URI\n- profile, key_prefix = get_db_info_from_uri(tracking_uri)\n- if profile:\n- # Reconstruct the profile string including key_prefix if present\n- profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n- return construct_db_uc_uri_from_profile(profile_string)\n- else:\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n-\n- # For non-databricks tracking URIs, use the tracking URI as the registry URI\n- return tracking_uri\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n-def _resolve_registry_uri(\n- registry_uri: str | None = None, tracking_uri: str | None = None\n-) -> str | None:\n- \"\"\"\n- Resolve the registry URI following the same logic as get_registry_uri().\n- \"\"\"\n- return (\n- registry_uri\n- or _get_registry_uri_from_context()\n- or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n-def _get_rest_store(store_uri, **_):\n- return RestStore(partial(get_default_host_creds, store_uri))\n+\n+\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n-def _get_store_registry():\n- global _model_registry_store_registry\n- from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n- from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n- if _model_registry_store_registry is not None:\n- return _model_registry_store_registry\n \n- _model_registry_store_registry = ModelRegistryStoreRegistry()\n- _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n- # Register a placeholder function that raises if users pass a registry URI with scheme\n- # \"databricks-uc\"\n- _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n- _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n- for scheme in [\"http\", \"https\"]:\n- _model_registry_store_registry.register(scheme, _get_rest_store)\n \n- for scheme in DATABASE_ENGINES:\n- _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n- for scheme in [\"\", \"file\"]:\n- _model_registry_store_registry.register(scheme, _get_file_store)\n \n- _model_registry_store_registry.register_entrypoints()\n- return _model_registry_store_registry\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n--- a/mlflow/tracking/_tracking_service/utils.py\n+++ b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n- @property\n- def tracking_uri(self):\n- return self._tracking_client.tracking_uri\n \n- def _get_registry_client(self):\n- \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n- Raises:\n- MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n- example, when the registry URI refers to an unsupported store type (e.g., the\n- FileStore).\n \n- Returns:\n- A ModelRegistryClient instance.\n- \"\"\"\n- # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n- # an instance variable on this `MlflowClient` instance. Because the instance variable\n- # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n- # function is used to safely fetch the variable (if it is defined) or a NoneType\n- # (if it is not defined)\n- registry_client_attr = \"_registry_client_lazy\"\n- registry_client = getattr(self, registry_client_attr, None)\n- if registry_client is None:\n- try:\n- registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n- # Define an instance variable on this `MlflowClient` instance to reference the\n- # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n- # that the variable name is consistent with the variable name specified in the\n- # preceding call to `getattr()`\n- setattr(self, registry_client_attr, registry_client)\n- except UnsupportedModelRegistryStoreURIException as exc:\n- raise MlflowException(\n- \"Model Registry features are not supported by the store with URI:\"\n- f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n- f\" {exc.supported_uri_schemes}.\",\n- FEATURE_DISABLED,\n- )\n- return registry_client\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEventStr | WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of events that trigger this webhook. Can be strings or\n- `WebhookEvent` objects.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self._get_registry_client().get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self._get_registry_client().list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEventStr | WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of events. Can be strings or `WebhookEvent` objects.\n- secret: New webhook secret.\n- status: New webhook status. Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().update_webhook(\n- webhook_id=webhook_id,\n- name=name,\n- description=description,\n- url=url,\n- events=events,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self._get_registry_client().delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n- return self._get_registry_client().test_webhook(webhook_id, event)\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,7 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,36 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n--- a/mlflow/utils/credentials.py\n+++ b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n-def get_default_host_creds(store_uri):\n- creds = read_mlflow_creds()\n- return MlflowHostCreds(\n- host=store_uri,\n- username=creds.username,\n- password=creds.password,\n- token=MLFLOW_TRACKING_TOKEN.get(),\n- aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n- auth=MLFLOW_TRACKING_AUTH.get(),\n- ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n- client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n- server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n-def parse_dict(js_dict, message):\n- \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n- _stringify_all_experiment_ids(js_dict)\n- ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n+\n+\n+\n+\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n--- a/mlflow/utils/request_utils.py\n+++ b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n-@lru_cache(maxsize=64)\n-def _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- # To create a new Session object for each process, we use the process id as the cache key.\n- # This is to avoid sharing the same Session object across processes, which can lead to issues\n- # such as https://stackoverflow.com/q/3724900.\n- _pid,\n- respect_retry_after_header=True,\n-):\n- \"\"\"\n- This function should not be called directly. Instead, use `_get_request_session` below.\n- \"\"\"\n \n- retry_kwargs = {\n- \"total\": max_retries,\n- \"connect\": max_retries,\n- \"read\": max_retries,\n- \"redirect\": max_retries,\n- \"status\": max_retries,\n- \"status_forcelist\": retry_codes,\n- \"backoff_factor\": backoff_factor,\n- \"backoff_jitter\": backoff_jitter,\n- \"raise_on_status\": raise_on_status,\n- \"respect_retry_after_header\": respect_retry_after_header,\n- }\n- urllib3_version = Version(urllib3.__version__)\n- if urllib3_version >= Version(\"1.26.0\"):\n- retry_kwargs[\"allowed_methods\"] = None\n- else:\n- retry_kwargs[\"method_whitelist\"] = None\n-\n- if urllib3_version < Version(\"2.0\"):\n- retry = JitteredRetry(**retry_kwargs)\n- else:\n- retry = Retry(**retry_kwargs)\n- from mlflow.environment_variables import (\n- MLFLOW_HTTP_POOL_CONNECTIONS,\n- MLFLOW_HTTP_POOL_MAXSIZE,\n- )\n-\n- adapter = HTTPAdapter(\n- pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n- pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n- max_retries=retry,\n- )\n- session = requests.Session()\n- session.mount(\"https://\", adapter)\n- session.mount(\"http://\", adapter)\n- return session\n-\n-\n-def _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n-):\n- \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n- Args:\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n- Returns:\n- requests.Session object.\n \n- \"\"\"\n- return _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- _pid=os.getpid(),\n- respect_retry_after_header=respect_retry_after_header,\n- )\n \n \n-def _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status=True,\n- allow_redirects=None,\n- respect_retry_after_header=True,\n- **kwargs,\n-):\n- \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n- Args:\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- url: The target URL address for the HTTP request.\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- session = _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n- )\n \n- # the environment variable is hardcoded here to avoid importing mlflow.\n- # however, documentation is available in environment_variables.py\n- env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n- allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n- return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n-from mlflow.utils.proto_json_utils import parse_dict\n-from mlflow.utils.request_utils import (\n- _TRANSIENT_FAILURE_RESPONSE_CODES,\n- _get_http_response_with_retries,\n- augmented_raise_for_status, # noqa: F401\n- cloud_storage_http_request, # noqa: F401\n-)\n-from mlflow.utils.string_utils import strip_suffix\n+\n+from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n+\n+\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n \n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n \n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n-\n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n \n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n-\n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n-\n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n-\n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n-\n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n-\n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n-\n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n-def _can_parse_as_json_object(string):\n- try:\n- return isinstance(json.loads(string), dict)\n- except Exception:\n- return False\n+\n+\n+\n+\n+\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n-def verify_rest_response(response, endpoint):\n- \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n- # Handle Armeria-specific response case where response text is \"200 OK\"\n- # v1/traces endpoint might return empty response\n- if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n- response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n- return response\n-\n- # Handle non-200 status codes\n- if response.status_code != 200:\n- if _can_parse_as_json_object(response.text):\n- raise RestException(json.loads(response.text))\n- else:\n- base_msg = (\n- f\"API request to endpoint {endpoint} \"\n- f\"failed with error code {response.status_code} != 200\"\n- )\n- raise MlflowException(\n- f\"{base_msg}. Response body: '{response.text}'\",\n- error_code=get_error_code(response.status_code),\n- )\n \n- # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n- # response\n- if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n- base_msg = (\n- \"API request to endpoint was successful but the response body was not \"\n- \"in a valid JSON format\"\n- )\n- raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n-\n- return response\n-\n-\n-def _validate_max_retries(max_retries):\n- max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n-\n- if max_retry_limit < 0:\n- raise MlflowException(\n- message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n- \"Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if max_retries >= max_retry_limit:\n- raise MlflowException(\n- message=f\"The configured max_retries value ({max_retries}) is \"\n- f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if max_retries < 0:\n- raise MlflowException(\n- message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n-\n-def _validate_backoff_factor(backoff_factor):\n- max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n-\n- if max_backoff_factor_limit < 0:\n- raise MlflowException(\n- message=\"The current maximum backoff factor limit is invalid \"\n- f\"({max_backoff_factor_limit}). Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor >= max_backoff_factor_limit:\n- raise MlflowException(\n- message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n- \"of the maximum allowable backoff_factor limit \"\n- f\"({max_backoff_factor_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor < 0:\n- raise MlflowException(\n- message=\"The backoff_factor value must be either 0 a positive integer. \"\n- f\"Got {backoff_factor}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n-def call_endpoint(\n- host_creds,\n- endpoint,\n- method,\n- json_body,\n- response_proto,\n- extra_headers=None,\n- retry_timeout_seconds=None,\n-):\n- # Convert json string to json dictionary, to pass to requests\n- if json_body is not None:\n- json_body = json.loads(json_body)\n- call_kwargs = {\n- \"host_creds\": host_creds,\n- \"endpoint\": endpoint,\n- \"method\": method,\n- }\n- if extra_headers is not None:\n- call_kwargs[\"extra_headers\"] = extra_headers\n- if retry_timeout_seconds is not None:\n- call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n- if method == \"GET\":\n- call_kwargs[\"params\"] = json_body\n- response = http_request(**call_kwargs)\n- else:\n- call_kwargs[\"json\"] = json_body\n- response = http_request(**call_kwargs)\n \n- response = verify_rest_response(response, endpoint)\n- response_to_parse = response.text\n- try:\n- js_dict = json.loads(response_to_parse)\n- except json.JSONDecodeError:\n- if len(response_to_parse) > 50:\n- response_to_parse = response_to_parse[:50] + \"...\"\n- _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n- raise\n \n- parse_dict(js_dict=js_dict, message=response_proto)\n- return response_proto\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n- def __init__(\n- self,\n- host,\n- username=None,\n- password=None,\n- token=None,\n- aws_sigv4=False,\n- auth=None,\n- ignore_tls_verification=False,\n- client_cert_path=None,\n- server_cert_path=None,\n- use_databricks_sdk=False,\n- databricks_auth_profile=None,\n- client_id=None,\n- client_secret=None,\n- use_secret_scope_token=False,\n- ):\n- if not host:\n- raise MlflowException(\n- message=\"host is a required parameter for MlflowHostCreds\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if ignore_tls_verification and (server_cert_path is not None):\n- raise MlflowException(\n- message=(\n- \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n- \"must not be set! This error may have occurred because the \"\n- \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n- \"environment variables are both set - only one of these environment \"\n- \"variables may be set.\"\n- ),\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- self.host = host\n- self.username = username\n- self.password = password\n- self.token = token\n- self.aws_sigv4 = aws_sigv4\n- self.auth = auth\n- self.ignore_tls_verification = ignore_tls_verification\n- self.client_cert_path = client_cert_path\n- self.server_cert_path = server_cert_path\n- self.use_databricks_sdk = use_databricks_sdk\n- self.databricks_auth_profile = databricks_auth_profile\n- self.client_id = client_id\n- self.client_secret = client_secret\n- self.use_secret_scope_token = use_secret_scope_token\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,12 +712,11 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n- @property\n- def verify(self):\n- if self.use_databricks_sdk:\n- # Let databricks-sdk set HTTP request `verify` param.\n- return None\n- if self.server_cert_path is None:\n- return not self.ignore_tls_verification\n- else:\n- return self.server_cert_path\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n+++ a/mlflow/entities/webhook.py\n--- b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n+ def __eq__(self, other: object) -> bool:\n+ if not isinstance(other, WebhookEvent):\n+ return False\n+ return self.entity == other.entity and self.action == other.action\n-\n-\n-\n-\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n+ @property\n+ def secret(self) -> str | None:\n+ return self._secret\n-\n-\n-\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n+++ a/mlflow/server/handlers.py\n--- b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n+++ a/mlflow/store/artifact/azure_blob_artifact_repo.py\n--- b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+from mlflow.utils.credentials import get_default_host_creds\n-\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ augmented_raise_for_status,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(self, store_uri=None, tracking_uri=None):\n+ \"\"\"\n+ Empty constructor. This is deliberately not marked as abstract, else every derived class\n+ would be forced to create one.\n \n+ Args:\n+ store_uri: The model registry store URI.\n+ tracking_uri: URI of the current MLflow tracking server, used to perform operations\n+ like fetching source run metadata or downloading source run artifacts\n+ to support subsequently uploading them to the model registry storage\n+ location.\n+ \"\"\"\n+ # Create a thread lock to ensure thread safety when linking prompts to other entities,\n+ # since the default linking implementation reads and appends entity tags, which\n+ # is prone to concurrent modification issues\n+ self._prompt_link_lock = threading.RLock()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n+++ a/mlflow/store/model_registry/base_rest_store.py\n--- b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n+from mlflow.utils.rest_utils import (\n+ call_endpoint,\n+ call_endpoints,\n+)\n-from mlflow.utils.rest_utils import call_endpoints\n-\n-\n-\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n+++ a/mlflow/tracking/_model_registry/client.py\n--- b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ @record_usage_event(CreateWebhookEvent)\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Unique name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of event types that trigger this webhook.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE).\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ return self.store.create_webhook(name, url, events, description, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self.store.get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self.store.list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of event types.\n+ secret: New webhook secret.\n+ status: New webhook status.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self.store.delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n+ return self.store.test_webhook(webhook_id, event)\n- return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n+++ a/mlflow/tracking/_model_registry/registry.py\n--- b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n+ from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n-\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n+++ a/mlflow/tracking/_model_registry/utils.py\n--- b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n+def _get_registry_uri_from_spark_session():\n+ session = _get_active_spark_session()\n+ if session is None:\n+ return None\n \n+ if is_in_databricks_serverless_runtime():\n+ # Connected to Serverless\n+ return \"databricks-uc\"\n \n+ from pyspark.sql.utils import AnalysisException\n \n+ try:\n+ return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n+ except AnalysisException:\n+ # In serverless clusters, session.conf.get() is unsupported\n+ # and raises an AnalysisException. We may encounter this case\n+ # when DBConnect is used to connect to a serverless cluster,\n+ # in which case the prior `is_in_databricks_serverless_runtime()`\n+ # check will have returned false (as of 2025-06-07, it checks\n+ # an environment variable that isn't set by DBConnect)\n+ return None\n \n \n+def _get_registry_uri_from_context():\n+ if _registry_uri is not None:\n+ return _registry_uri\n+ elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n+ return uri\n+ return _registry_uri\n \n \n+def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n+ \"\"\"\n+ Get the default registry URI for a given tracking URI.\n \n+ If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n+ Otherwise, returns the tracking URI itself.\n \n+ Args:\n+ tracking_uri: The tracking URI to get the default registry URI for\n \n+ Returns:\n+ The default registry URI\n+ \"\"\"\n+ if tracking_uri is not None and is_databricks_uri(tracking_uri):\n+ # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n+ # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n+ # model registry offering on Databricks\n+ if tracking_uri == \"databricks\":\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+ else:\n+ # Extract profile from tracking URI and construct databricks-uc URI\n+ profile, key_prefix = get_db_info_from_uri(tracking_uri)\n+ if profile:\n+ # Reconstruct the profile string including key_prefix if present\n+ profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n+ return construct_db_uc_uri_from_profile(profile_string)\n+ else:\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+\n+ # For non-databricks tracking URIs, use the tracking URI as the registry URI\n+ return tracking_uri\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n+def _resolve_registry_uri(\n+ registry_uri: str | None = None, tracking_uri: str | None = None\n+) -> str | None:\n+ \"\"\"\n+ Resolve the registry URI following the same logic as get_registry_uri().\n+ \"\"\"\n+ return (\n+ registry_uri\n+ or _get_registry_uri_from_context()\n+ or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n+def _get_rest_store(store_uri, **_):\n+ return RestStore(partial(get_default_host_creds, store_uri))\n-\n-\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n+def _get_store_registry():\n+ global _model_registry_store_registry\n+ from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n+ from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n+ if _model_registry_store_registry is not None:\n+ return _model_registry_store_registry\n \n+ _model_registry_store_registry = ModelRegistryStoreRegistry()\n+ _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n+ # Register a placeholder function that raises if users pass a registry URI with scheme\n+ # \"databricks-uc\"\n+ _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n+ _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n+ for scheme in [\"http\", \"https\"]:\n+ _model_registry_store_registry.register(scheme, _get_rest_store)\n \n+ for scheme in DATABASE_ENGINES:\n+ _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n+ for scheme in [\"\", \"file\"]:\n+ _model_registry_store_registry.register(scheme, _get_file_store)\n \n+ _model_registry_store_registry.register_entrypoints()\n+ return _model_registry_store_registry\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n+++ a/mlflow/tracking/_tracking_service/utils.py\n--- b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n+++ a/mlflow/tracking/client.py\n--- b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n+ @property\n+ def tracking_uri(self):\n+ return self._tracking_client.tracking_uri\n \n+ def _get_registry_client(self):\n+ \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n+ Raises:\n+ MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n+ example, when the registry URI refers to an unsupported store type (e.g., the\n+ FileStore).\n \n+ Returns:\n+ A ModelRegistryClient instance.\n+ \"\"\"\n+ # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n+ # an instance variable on this `MlflowClient` instance. Because the instance variable\n+ # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n+ # function is used to safely fetch the variable (if it is defined) or a NoneType\n+ # (if it is not defined)\n+ registry_client_attr = \"_registry_client_lazy\"\n+ registry_client = getattr(self, registry_client_attr, None)\n+ if registry_client is None:\n+ try:\n+ registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n+ # Define an instance variable on this `MlflowClient` instance to reference the\n+ # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n+ # that the variable name is consistent with the variable name specified in the\n+ # preceding call to `getattr()`\n+ setattr(self, registry_client_attr, registry_client)\n+ except UnsupportedModelRegistryStoreURIException as exc:\n+ raise MlflowException(\n+ \"Model Registry features are not supported by the store with URI:\"\n+ f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n+ f\" {exc.supported_uri_schemes}.\",\n+ FEATURE_DISABLED,\n+ )\n+ return registry_client\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEventStr | WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of events that trigger this webhook. Can be strings or\n+ `WebhookEvent` objects.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self._get_registry_client().get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self._get_registry_client().list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEventStr | WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of events. Can be strings or `WebhookEvent` objects.\n+ secret: New webhook secret.\n+ status: New webhook status. Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ if events is not None:\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().update_webhook(\n+ webhook_id=webhook_id,\n+ name=name,\n+ description=description,\n+ url=url,\n+ events=events,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self._get_registry_client().delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n- return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,7 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,36 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n+++ a/mlflow/utils/credentials.py\n--- b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n+def get_default_host_creds(store_uri):\n+ creds = read_mlflow_creds()\n+ return MlflowHostCreds(\n+ host=store_uri,\n+ username=creds.username,\n+ password=creds.password,\n+ token=MLFLOW_TRACKING_TOKEN.get(),\n+ aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n+ auth=MLFLOW_TRACKING_AUTH.get(),\n+ ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n+ client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n+ server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n+def parse_dict(js_dict, message):\n+ \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n+ _stringify_all_experiment_ids(js_dict)\n+ ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n-\n-\n-\n-\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n+++ a/mlflow/utils/request_utils.py\n--- b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n+@lru_cache(maxsize=64)\n+def _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ # To create a new Session object for each process, we use the process id as the cache key.\n+ # This is to avoid sharing the same Session object across processes, which can lead to issues\n+ # such as https://stackoverflow.com/q/3724900.\n+ _pid,\n+ respect_retry_after_header=True,\n+):\n+ \"\"\"\n+ This function should not be called directly. Instead, use `_get_request_session` below.\n+ \"\"\"\n \n+ retry_kwargs = {\n+ \"total\": max_retries,\n+ \"connect\": max_retries,\n+ \"read\": max_retries,\n+ \"redirect\": max_retries,\n+ \"status\": max_retries,\n+ \"status_forcelist\": retry_codes,\n+ \"backoff_factor\": backoff_factor,\n+ \"backoff_jitter\": backoff_jitter,\n+ \"raise_on_status\": raise_on_status,\n+ \"respect_retry_after_header\": respect_retry_after_header,\n+ }\n+ urllib3_version = Version(urllib3.__version__)\n+ if urllib3_version >= Version(\"1.26.0\"):\n+ retry_kwargs[\"allowed_methods\"] = None\n+ else:\n+ retry_kwargs[\"method_whitelist\"] = None\n+\n+ if urllib3_version < Version(\"2.0\"):\n+ retry = JitteredRetry(**retry_kwargs)\n+ else:\n+ retry = Retry(**retry_kwargs)\n+ from mlflow.environment_variables import (\n+ MLFLOW_HTTP_POOL_CONNECTIONS,\n+ MLFLOW_HTTP_POOL_MAXSIZE,\n+ )\n+\n+ adapter = HTTPAdapter(\n+ pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n+ pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n+ max_retries=retry,\n+ )\n+ session = requests.Session()\n+ session.mount(\"https://\", adapter)\n+ session.mount(\"http://\", adapter)\n+ return session\n+\n+\n+def _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+):\n+ \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n+ Args:\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n+ Returns:\n+ requests.Session object.\n \n+ \"\"\"\n+ return _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ _pid=os.getpid(),\n+ respect_retry_after_header=respect_retry_after_header,\n+ )\n \n \n+def _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status=True,\n+ allow_redirects=None,\n+ respect_retry_after_header=True,\n+ **kwargs,\n+):\n+ \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n+ Args:\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ url: The target URL address for the HTTP request.\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ session = _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+ )\n \n+ # the environment variable is hardcoded here to avoid importing mlflow.\n+ # however, documentation is available in environment_variables.py\n+ env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n+ allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n+ return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n+from mlflow.utils.proto_json_utils import parse_dict\n+from mlflow.utils.request_utils import (\n+ _TRANSIENT_FAILURE_RESPONSE_CODES,\n+ _get_http_response_with_retries,\n+ augmented_raise_for_status, # noqa: F401\n+ cloud_storage_http_request, # noqa: F401\n+)\n+from mlflow.utils.string_utils import strip_suffix\n-\n-from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n-\n-\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n \n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n \n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n+\n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n \n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n+\n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n+\n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n+\n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n+\n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n+\n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n+\n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n+def _can_parse_as_json_object(string):\n+ try:\n+ return isinstance(json.loads(string), dict)\n+ except Exception:\n+ return False\n-\n-\n-\n-\n-\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n+def verify_rest_response(response, endpoint):\n+ \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n+ # Handle Armeria-specific response case where response text is \"200 OK\"\n+ # v1/traces endpoint might return empty response\n+ if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n+ response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n+ return response\n+\n+ # Handle non-200 status codes\n+ if response.status_code != 200:\n+ if _can_parse_as_json_object(response.text):\n+ raise RestException(json.loads(response.text))\n+ else:\n+ base_msg = (\n+ f\"API request to endpoint {endpoint} \"\n+ f\"failed with error code {response.status_code} != 200\"\n+ )\n+ raise MlflowException(\n+ f\"{base_msg}. Response body: '{response.text}'\",\n+ error_code=get_error_code(response.status_code),\n+ )\n \n+ # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n+ # response\n+ if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n+ base_msg = (\n+ \"API request to endpoint was successful but the response body was not \"\n+ \"in a valid JSON format\"\n+ )\n+ raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n+\n+ return response\n+\n+\n+def _validate_max_retries(max_retries):\n+ max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n+\n+ if max_retry_limit < 0:\n+ raise MlflowException(\n+ message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n+ \"Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if max_retries >= max_retry_limit:\n+ raise MlflowException(\n+ message=f\"The configured max_retries value ({max_retries}) is \"\n+ f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if max_retries < 0:\n+ raise MlflowException(\n+ message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+\n+def _validate_backoff_factor(backoff_factor):\n+ max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n+\n+ if max_backoff_factor_limit < 0:\n+ raise MlflowException(\n+ message=\"The current maximum backoff factor limit is invalid \"\n+ f\"({max_backoff_factor_limit}). Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor >= max_backoff_factor_limit:\n+ raise MlflowException(\n+ message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n+ \"of the maximum allowable backoff_factor limit \"\n+ f\"({max_backoff_factor_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor < 0:\n+ raise MlflowException(\n+ message=\"The backoff_factor value must be either 0 a positive integer. \"\n+ f\"Got {backoff_factor}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n+def call_endpoint(\n+ host_creds,\n+ endpoint,\n+ method,\n+ json_body,\n+ response_proto,\n+ extra_headers=None,\n+ retry_timeout_seconds=None,\n+):\n+ # Convert json string to json dictionary, to pass to requests\n+ if json_body is not None:\n+ json_body = json.loads(json_body)\n+ call_kwargs = {\n+ \"host_creds\": host_creds,\n+ \"endpoint\": endpoint,\n+ \"method\": method,\n+ }\n+ if extra_headers is not None:\n+ call_kwargs[\"extra_headers\"] = extra_headers\n+ if retry_timeout_seconds is not None:\n+ call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n+ if method == \"GET\":\n+ call_kwargs[\"params\"] = json_body\n+ response = http_request(**call_kwargs)\n+ else:\n+ call_kwargs[\"json\"] = json_body\n+ response = http_request(**call_kwargs)\n \n+ response = verify_rest_response(response, endpoint)\n+ response_to_parse = response.text\n+ try:\n+ js_dict = json.loads(response_to_parse)\n+ except json.JSONDecodeError:\n+ if len(response_to_parse) > 50:\n+ response_to_parse = response_to_parse[:50] + \"...\"\n+ _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n+ raise\n \n+ parse_dict(js_dict=js_dict, message=response_proto)\n+ return response_proto\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n+ def __init__(\n+ self,\n+ host,\n+ username=None,\n+ password=None,\n+ token=None,\n+ aws_sigv4=False,\n+ auth=None,\n+ ignore_tls_verification=False,\n+ client_cert_path=None,\n+ server_cert_path=None,\n+ use_databricks_sdk=False,\n+ databricks_auth_profile=None,\n+ client_id=None,\n+ client_secret=None,\n+ use_secret_scope_token=False,\n+ ):\n+ if not host:\n+ raise MlflowException(\n+ message=\"host is a required parameter for MlflowHostCreds\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if ignore_tls_verification and (server_cert_path is not None):\n+ raise MlflowException(\n+ message=(\n+ \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n+ \"must not be set! This error may have occurred because the \"\n+ \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n+ \"environment variables are both set - only one of these environment \"\n+ \"variables may be set.\"\n+ ),\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ self.host = host\n+ self.username = username\n+ self.password = password\n+ self.token = token\n+ self.aws_sigv4 = aws_sigv4\n+ self.auth = auth\n+ self.ignore_tls_verification = ignore_tls_verification\n+ self.client_cert_path = client_cert_path\n+ self.server_cert_path = server_cert_path\n+ self.use_databricks_sdk = use_databricks_sdk\n+ self.databricks_auth_profile = databricks_auth_profile\n+ self.client_id = client_id\n+ self.client_secret = client_secret\n+ self.use_secret_scope_token = use_secret_scope_token\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,11 +712,12 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n+ @property\n+ def verify(self):\n+ if self.use_databricks_sdk:\n+ # Let databricks-sdk set HTTP request `verify` param.\n+ return None\n+ if self.server_cert_path is None:\n+ return not self.ignore_tls_verification\n+ else:\n+ return self.server_cert_path\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..e8ca4a509 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -6405,7 +6405,15 @@ class MlflowClient:\n A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n \"\"\"\n if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ converted_events = []\n+ for e in events:\n+ if isinstance(e, str):\n+ try:\n+ e = WebhookEvent.from_str(e)\n+ except Exception:\n+ continue\n+ converted_events.append(e)\n+ events = converted_events\n \n if status is not None:\n status = WebhookStatus(status) if isinstance(status, str) else status\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py\ndeleted file mode 100644\nindex 97fd61b00..000000000\n--- a/tests/tracking/test_client_webhooks.py\n+++ /dev/null\n@@ -1,219 +0,0 @@\n-from pathlib import Path\n-from typing import Iterator\n-\n-import pytest\n-from cryptography.fernet import Fernet\n-\n-from mlflow.entities.webhook import WebhookAction, WebhookEntity, WebhookEvent, WebhookStatus\n-from mlflow.environment_variables import MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY\n-from mlflow.exceptions import MlflowException\n-from mlflow.server import handlers\n-from mlflow.server.fastapi_app import app\n-from mlflow.server.handlers import initialize_backend_stores\n-from mlflow.tracking import MlflowClient\n-\n-from tests.helper_functions import get_safe_port\n-from tests.tracking.integration_test_utils import ServerThread\n-\n-\n-@pytest.fixture\n-def client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Iterator[MlflowClient]:\n- \"\"\"Setup a local MLflow server with proper webhook encryption key support.\"\"\"\n- # Set up encryption key for webhooks using monkeypatch\n- encryption_key = Fernet.generate_key().decode(\"utf-8\")\n- monkeypatch.setenv(MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY.name, encryption_key)\n-\n- # Configure backend stores\n- backend_uri = f\"sqlite:///{tmp_path / 'mlflow.db'}\"\n- default_artifact_root = tmp_path.as_uri()\n-\n- # Force-reset backend stores before each test\n- handlers._tracking_store = None\n- handlers._model_registry_store = None\n- initialize_backend_stores(backend_uri, default_artifact_root=default_artifact_root)\n-\n- # Start server and return client\n- with ServerThread(app, get_safe_port()) as url:\n- yield MlflowClient(url)\n-\n-\n-def test_create_webhook(client: MlflowClient):\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n- assert webhook.events == [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n-\n- webhook = client.get_webhook(webhook.webhook_id)\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n-\n- # With secret\n- webhook_with_secret = client.create_webhook(\n- name=\"test_webhook_with_secret\",\n- url=\"https://example.com/webhook_with_secret\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- secret=\"my_secret\",\n- )\n- assert webhook_with_secret.name == \"test_webhook_with_secret\"\n- assert webhook_with_secret.url == \"https://example.com/webhook_with_secret\"\n- assert webhook_with_secret.secret is None\n- assert webhook_with_secret.events == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)\n- ]\n-\n- # Multiple events\n- webhook_multiple_events = client.create_webhook(\n- name=\"test_webhook_multiple_events\",\n- url=\"https://example.com/webhook_multiple_events\",\n- events=[\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- ],\n- )\n- assert webhook_multiple_events.name == \"test_webhook_multiple_events\"\n- assert webhook_multiple_events.url == \"https://example.com/webhook_multiple_events\"\n- assert sorted(\n- webhook_multiple_events.events, key=lambda e: (e.entity.value, e.action.value)\n- ) == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- ]\n- assert webhook_multiple_events.secret is None\n-\n-\n-def test_get_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- created_webhook = client.create_webhook(\n- name=\"test_webhook\", url=\"https://example.com/webhook\", events=events\n- )\n- retrieved_webhook = client.get_webhook(created_webhook.webhook_id)\n- assert retrieved_webhook.webhook_id == created_webhook.webhook_id\n- assert retrieved_webhook.name == \"test_webhook\"\n- assert retrieved_webhook.url == \"https://example.com/webhook\"\n- assert retrieved_webhook.events == events\n-\n-\n-def test_get_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.get_webhook(\"nonexistent\")\n-\n-\n-def test_list_webhooks(client: MlflowClient):\n- # Create more webhooks than max_results\n- created_webhooks = []\n- for i in range(5):\n- webhook = client.create_webhook(\n- name=f\"webhook{i}\",\n- url=f\"https://example.com/{i}\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- created_webhooks.append(webhook)\n- # Test pagination with max_results=2\n- webhooks_page = client.list_webhooks(max_results=2)\n- assert len(webhooks_page) == 2\n- assert webhooks_page.token is not None\n- # Get next page\n- next_webhooks_page = client.list_webhooks(max_results=2, page_token=webhooks_page.token)\n- assert len(next_webhooks_page) == 2\n- assert next_webhooks_page.token is not None\n- # Verify we don't get duplicates\n- first_page_ids = {w.webhook_id for w in webhooks_page}\n- second_page_ids = {w.webhook_id for w in next_webhooks_page}\n- assert first_page_ids.isdisjoint(second_page_ids)\n-\n-\n-def test_update_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\", url=\"https://example.com/original\", events=events\n- )\n- # Update webhook\n- new_events = [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.REGISTERED_MODEL, WebhookAction.CREATED),\n- ]\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- url=\"https://example.com/updated\",\n- events=new_events,\n- description=\"Updated description\",\n- secret=\"new_secret\",\n- status=WebhookStatus.DISABLED,\n- )\n- assert updated_webhook.webhook_id == webhook.webhook_id\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/updated\"\n- assert updated_webhook.events == new_events\n- assert updated_webhook.description == \"Updated description\"\n- assert updated_webhook.status == WebhookStatus.DISABLED\n- assert updated_webhook.last_updated_timestamp > webhook.last_updated_timestamp\n-\n-\n-def test_update_webhook_partial(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\",\n- url=\"https://example.com/original\",\n- events=events,\n- description=\"Original description\",\n- )\n- # Update only the name\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- )\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/original\"\n- assert updated_webhook.events == events\n- assert updated_webhook.description == \"Original description\"\n-\n-\n-def test_update_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.update_webhook(webhook_id=\"nonexistent\", name=\"new_name\")\n-\n-\n-@pytest.mark.parametrize(\n- (\"invalid_url\", \"expected_match\"),\n- [\n- (\" \", r\"Webhook URL cannot be empty or just whitespace\"),\n- (\"ftp://example.com\", r\"Invalid webhook URL scheme\"),\n- (\"http://[invalid\", r\"Invalid webhook URL\"),\n- ],\n-)\n-def test_update_webhook_invalid_urls(client: MlflowClient, invalid_url: str, expected_match: str):\n- # Create a valid webhook first\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- with pytest.raises(MlflowException, match=expected_match):\n- client.update_webhook(webhook_id=webhook.webhook_id, url=invalid_url)\n-\n-\n-def test_delete_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=events,\n- )\n- client.delete_webhook(webhook.webhook_id)\n- with pytest.raises(MlflowException, match=r\"Webhook with ID .* not found\"):\n- client.get_webhook(webhook.webhook_id)\n- webhooks_page = client.list_webhooks()\n- webhook_ids = {w.webhook_id for w in webhooks_page}\n- assert webhook.webhook_id not in webhook_ids\n-\n-\n-def test_delete_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.delete_webhook(\"nonexistent\")\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S8\",\n \"shared_functions\": 2,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S9\",\n \"shared_functions\": 2,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":61,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S13\",\n \"text\": \"Delete a webhook. This method removes a webhook from the MLflow Model Registry. Once deleted, the webhook will no longer receive notifications for the events it was configured to monitor. Signature: delete_webhook(self, webhook_id: str) -> None. Args: webhook_id: The unique identifier of the webhook to delete. This ID is returned when creating a webhook or can be obtained from listing webhooks. Returns: None. Task statement: Create, retrieve, update, and delete webhooks for MLflow model registry events.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracking/client.py\",\n \"lines\": \"6427-6437\",\n \"function\": \"MlflowClient.delete_webhook\"\n },\n {\n \"file\": \"mlflow/tracking/_model_registry/client.py\",\n \"lines\": \"884-895\",\n \"function\": \"ModelRegistryClient.delete_webhook\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n--- a/mlflow/entities/webhook.py\n+++ b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n- def __eq__(self, other: object) -> bool:\n- if not isinstance(other, WebhookEvent):\n- return False\n- return self.entity == other.entity and self.action == other.action\n+\n+\n+\n+\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n- @property\n- def secret(self) -> str | None:\n- return self._secret\n+\n+\n+\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n--- a/mlflow/server/handlers.py\n+++ b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n--- a/mlflow/store/artifact/azure_blob_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-from mlflow.utils.credentials import get_default_host_creds\n+\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- augmented_raise_for_status,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n- def __init__(self, store_uri=None, tracking_uri=None):\n- \"\"\"\n- Empty constructor. This is deliberately not marked as abstract, else every derived class\n- would be forced to create one.\n \n- Args:\n- store_uri: The model registry store URI.\n- tracking_uri: URI of the current MLflow tracking server, used to perform operations\n- like fetching source run metadata or downloading source run artifacts\n- to support subsequently uploading them to the model registry storage\n- location.\n- \"\"\"\n- # Create a thread lock to ensure thread safety when linking prompts to other entities,\n- # since the default linking implementation reads and appends entity tags, which\n- # is prone to concurrent modification issues\n- self._prompt_link_lock = threading.RLock()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n--- a/mlflow/store/model_registry/base_rest_store.py\n+++ b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n-from mlflow.utils.rest_utils import (\n- call_endpoint,\n- call_endpoints,\n-)\n+from mlflow.utils.rest_utils import call_endpoints\n+\n+\n+\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- @record_usage_event(CreateWebhookEvent)\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Unique name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of event types that trigger this webhook.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE).\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- return self.store.create_webhook(name, url, events, description, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self.store.get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self.store.list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of event types.\n- secret: New webhook secret.\n- status: New webhook status.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self.store.delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n- return self.store.test_webhook(webhook_id, event)\n+ return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n--- a/mlflow/tracking/_model_registry/registry.py\n+++ b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n- from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n+\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n--- a/mlflow/tracking/_model_registry/utils.py\n+++ b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n-def _get_registry_uri_from_spark_session():\n- session = _get_active_spark_session()\n- if session is None:\n- return None\n \n- if is_in_databricks_serverless_runtime():\n- # Connected to Serverless\n- return \"databricks-uc\"\n \n- from pyspark.sql.utils import AnalysisException\n \n- try:\n- return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n- except AnalysisException:\n- # In serverless clusters, session.conf.get() is unsupported\n- # and raises an AnalysisException. We may encounter this case\n- # when DBConnect is used to connect to a serverless cluster,\n- # in which case the prior `is_in_databricks_serverless_runtime()`\n- # check will have returned false (as of 2025-06-07, it checks\n- # an environment variable that isn't set by DBConnect)\n- return None\n \n \n-def _get_registry_uri_from_context():\n- if _registry_uri is not None:\n- return _registry_uri\n- elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n- return uri\n- return _registry_uri\n \n \n-def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n- \"\"\"\n- Get the default registry URI for a given tracking URI.\n \n- If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n- Otherwise, returns the tracking URI itself.\n \n- Args:\n- tracking_uri: The tracking URI to get the default registry URI for\n \n- Returns:\n- The default registry URI\n- \"\"\"\n- if tracking_uri is not None and is_databricks_uri(tracking_uri):\n- # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n- # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n- # model registry offering on Databricks\n- if tracking_uri == \"databricks\":\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n- else:\n- # Extract profile from tracking URI and construct databricks-uc URI\n- profile, key_prefix = get_db_info_from_uri(tracking_uri)\n- if profile:\n- # Reconstruct the profile string including key_prefix if present\n- profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n- return construct_db_uc_uri_from_profile(profile_string)\n- else:\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n-\n- # For non-databricks tracking URIs, use the tracking URI as the registry URI\n- return tracking_uri\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n-def _resolve_registry_uri(\n- registry_uri: str | None = None, tracking_uri: str | None = None\n-) -> str | None:\n- \"\"\"\n- Resolve the registry URI following the same logic as get_registry_uri().\n- \"\"\"\n- return (\n- registry_uri\n- or _get_registry_uri_from_context()\n- or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n-def _get_rest_store(store_uri, **_):\n- return RestStore(partial(get_default_host_creds, store_uri))\n+\n+\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n-def _get_store_registry():\n- global _model_registry_store_registry\n- from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n- from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n- if _model_registry_store_registry is not None:\n- return _model_registry_store_registry\n \n- _model_registry_store_registry = ModelRegistryStoreRegistry()\n- _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n- # Register a placeholder function that raises if users pass a registry URI with scheme\n- # \"databricks-uc\"\n- _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n- _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n- for scheme in [\"http\", \"https\"]:\n- _model_registry_store_registry.register(scheme, _get_rest_store)\n \n- for scheme in DATABASE_ENGINES:\n- _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n- for scheme in [\"\", \"file\"]:\n- _model_registry_store_registry.register(scheme, _get_file_store)\n \n- _model_registry_store_registry.register_entrypoints()\n- return _model_registry_store_registry\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n--- a/mlflow/tracking/_tracking_service/utils.py\n+++ b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n- @property\n- def tracking_uri(self):\n- return self._tracking_client.tracking_uri\n \n- def _get_registry_client(self):\n- \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n- Raises:\n- MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n- example, when the registry URI refers to an unsupported store type (e.g., the\n- FileStore).\n \n- Returns:\n- A ModelRegistryClient instance.\n- \"\"\"\n- # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n- # an instance variable on this `MlflowClient` instance. Because the instance variable\n- # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n- # function is used to safely fetch the variable (if it is defined) or a NoneType\n- # (if it is not defined)\n- registry_client_attr = \"_registry_client_lazy\"\n- registry_client = getattr(self, registry_client_attr, None)\n- if registry_client is None:\n- try:\n- registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n- # Define an instance variable on this `MlflowClient` instance to reference the\n- # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n- # that the variable name is consistent with the variable name specified in the\n- # preceding call to `getattr()`\n- setattr(self, registry_client_attr, registry_client)\n- except UnsupportedModelRegistryStoreURIException as exc:\n- raise MlflowException(\n- \"Model Registry features are not supported by the store with URI:\"\n- f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n- f\" {exc.supported_uri_schemes}.\",\n- FEATURE_DISABLED,\n- )\n- return registry_client\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEventStr | WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of events that trigger this webhook. Can be strings or\n- `WebhookEvent` objects.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self._get_registry_client().get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self._get_registry_client().list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEventStr | WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of events. Can be strings or `WebhookEvent` objects.\n- secret: New webhook secret.\n- status: New webhook status. Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().update_webhook(\n- webhook_id=webhook_id,\n- name=name,\n- description=description,\n- url=url,\n- events=events,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self._get_registry_client().delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n- return self._get_registry_client().test_webhook(webhook_id, event)\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,7 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,36 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n--- a/mlflow/utils/credentials.py\n+++ b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n-def get_default_host_creds(store_uri):\n- creds = read_mlflow_creds()\n- return MlflowHostCreds(\n- host=store_uri,\n- username=creds.username,\n- password=creds.password,\n- token=MLFLOW_TRACKING_TOKEN.get(),\n- aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n- auth=MLFLOW_TRACKING_AUTH.get(),\n- ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n- client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n- server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n-def parse_dict(js_dict, message):\n- \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n- _stringify_all_experiment_ids(js_dict)\n- ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n+\n+\n+\n+\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n--- a/mlflow/utils/request_utils.py\n+++ b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n-@lru_cache(maxsize=64)\n-def _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- # To create a new Session object for each process, we use the process id as the cache key.\n- # This is to avoid sharing the same Session object across processes, which can lead to issues\n- # such as https://stackoverflow.com/q/3724900.\n- _pid,\n- respect_retry_after_header=True,\n-):\n- \"\"\"\n- This function should not be called directly. Instead, use `_get_request_session` below.\n- \"\"\"\n \n- retry_kwargs = {\n- \"total\": max_retries,\n- \"connect\": max_retries,\n- \"read\": max_retries,\n- \"redirect\": max_retries,\n- \"status\": max_retries,\n- \"status_forcelist\": retry_codes,\n- \"backoff_factor\": backoff_factor,\n- \"backoff_jitter\": backoff_jitter,\n- \"raise_on_status\": raise_on_status,\n- \"respect_retry_after_header\": respect_retry_after_header,\n- }\n- urllib3_version = Version(urllib3.__version__)\n- if urllib3_version >= Version(\"1.26.0\"):\n- retry_kwargs[\"allowed_methods\"] = None\n- else:\n- retry_kwargs[\"method_whitelist\"] = None\n-\n- if urllib3_version < Version(\"2.0\"):\n- retry = JitteredRetry(**retry_kwargs)\n- else:\n- retry = Retry(**retry_kwargs)\n- from mlflow.environment_variables import (\n- MLFLOW_HTTP_POOL_CONNECTIONS,\n- MLFLOW_HTTP_POOL_MAXSIZE,\n- )\n-\n- adapter = HTTPAdapter(\n- pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n- pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n- max_retries=retry,\n- )\n- session = requests.Session()\n- session.mount(\"https://\", adapter)\n- session.mount(\"http://\", adapter)\n- return session\n-\n-\n-def _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n-):\n- \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n- Args:\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n- Returns:\n- requests.Session object.\n \n- \"\"\"\n- return _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- _pid=os.getpid(),\n- respect_retry_after_header=respect_retry_after_header,\n- )\n \n \n-def _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status=True,\n- allow_redirects=None,\n- respect_retry_after_header=True,\n- **kwargs,\n-):\n- \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n- Args:\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- url: The target URL address for the HTTP request.\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- session = _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n- )\n \n- # the environment variable is hardcoded here to avoid importing mlflow.\n- # however, documentation is available in environment_variables.py\n- env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n- allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n- return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n-from mlflow.utils.proto_json_utils import parse_dict\n-from mlflow.utils.request_utils import (\n- _TRANSIENT_FAILURE_RESPONSE_CODES,\n- _get_http_response_with_retries,\n- augmented_raise_for_status, # noqa: F401\n- cloud_storage_http_request, # noqa: F401\n-)\n-from mlflow.utils.string_utils import strip_suffix\n+\n+from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n+\n+\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n \n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n \n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n-\n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n \n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n-\n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n-\n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n-\n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n-\n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n-\n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n-\n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n-def _can_parse_as_json_object(string):\n- try:\n- return isinstance(json.loads(string), dict)\n- except Exception:\n- return False\n+\n+\n+\n+\n+\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n-def verify_rest_response(response, endpoint):\n- \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n- # Handle Armeria-specific response case where response text is \"200 OK\"\n- # v1/traces endpoint might return empty response\n- if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n- response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n- return response\n-\n- # Handle non-200 status codes\n- if response.status_code != 200:\n- if _can_parse_as_json_object(response.text):\n- raise RestException(json.loads(response.text))\n- else:\n- base_msg = (\n- f\"API request to endpoint {endpoint} \"\n- f\"failed with error code {response.status_code} != 200\"\n- )\n- raise MlflowException(\n- f\"{base_msg}. Response body: '{response.text}'\",\n- error_code=get_error_code(response.status_code),\n- )\n \n- # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n- # response\n- if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n- base_msg = (\n- \"API request to endpoint was successful but the response body was not \"\n- \"in a valid JSON format\"\n- )\n- raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n-\n- return response\n-\n-\n-def _validate_max_retries(max_retries):\n- max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n-\n- if max_retry_limit < 0:\n- raise MlflowException(\n- message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n- \"Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if max_retries >= max_retry_limit:\n- raise MlflowException(\n- message=f\"The configured max_retries value ({max_retries}) is \"\n- f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if max_retries < 0:\n- raise MlflowException(\n- message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n-\n-def _validate_backoff_factor(backoff_factor):\n- max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n-\n- if max_backoff_factor_limit < 0:\n- raise MlflowException(\n- message=\"The current maximum backoff factor limit is invalid \"\n- f\"({max_backoff_factor_limit}). Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor >= max_backoff_factor_limit:\n- raise MlflowException(\n- message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n- \"of the maximum allowable backoff_factor limit \"\n- f\"({max_backoff_factor_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor < 0:\n- raise MlflowException(\n- message=\"The backoff_factor value must be either 0 a positive integer. \"\n- f\"Got {backoff_factor}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n-def call_endpoint(\n- host_creds,\n- endpoint,\n- method,\n- json_body,\n- response_proto,\n- extra_headers=None,\n- retry_timeout_seconds=None,\n-):\n- # Convert json string to json dictionary, to pass to requests\n- if json_body is not None:\n- json_body = json.loads(json_body)\n- call_kwargs = {\n- \"host_creds\": host_creds,\n- \"endpoint\": endpoint,\n- \"method\": method,\n- }\n- if extra_headers is not None:\n- call_kwargs[\"extra_headers\"] = extra_headers\n- if retry_timeout_seconds is not None:\n- call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n- if method == \"GET\":\n- call_kwargs[\"params\"] = json_body\n- response = http_request(**call_kwargs)\n- else:\n- call_kwargs[\"json\"] = json_body\n- response = http_request(**call_kwargs)\n \n- response = verify_rest_response(response, endpoint)\n- response_to_parse = response.text\n- try:\n- js_dict = json.loads(response_to_parse)\n- except json.JSONDecodeError:\n- if len(response_to_parse) > 50:\n- response_to_parse = response_to_parse[:50] + \"...\"\n- _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n- raise\n \n- parse_dict(js_dict=js_dict, message=response_proto)\n- return response_proto\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n- def __init__(\n- self,\n- host,\n- username=None,\n- password=None,\n- token=None,\n- aws_sigv4=False,\n- auth=None,\n- ignore_tls_verification=False,\n- client_cert_path=None,\n- server_cert_path=None,\n- use_databricks_sdk=False,\n- databricks_auth_profile=None,\n- client_id=None,\n- client_secret=None,\n- use_secret_scope_token=False,\n- ):\n- if not host:\n- raise MlflowException(\n- message=\"host is a required parameter for MlflowHostCreds\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if ignore_tls_verification and (server_cert_path is not None):\n- raise MlflowException(\n- message=(\n- \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n- \"must not be set! This error may have occurred because the \"\n- \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n- \"environment variables are both set - only one of these environment \"\n- \"variables may be set.\"\n- ),\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- self.host = host\n- self.username = username\n- self.password = password\n- self.token = token\n- self.aws_sigv4 = aws_sigv4\n- self.auth = auth\n- self.ignore_tls_verification = ignore_tls_verification\n- self.client_cert_path = client_cert_path\n- self.server_cert_path = server_cert_path\n- self.use_databricks_sdk = use_databricks_sdk\n- self.databricks_auth_profile = databricks_auth_profile\n- self.client_id = client_id\n- self.client_secret = client_secret\n- self.use_secret_scope_token = use_secret_scope_token\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,12 +712,11 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n- @property\n- def verify(self):\n- if self.use_databricks_sdk:\n- # Let databricks-sdk set HTTP request `verify` param.\n- return None\n- if self.server_cert_path is None:\n- return not self.ignore_tls_verification\n- else:\n- return self.server_cert_path\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n+++ a/mlflow/entities/webhook.py\n--- b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n+ def __eq__(self, other: object) -> bool:\n+ if not isinstance(other, WebhookEvent):\n+ return False\n+ return self.entity == other.entity and self.action == other.action\n-\n-\n-\n-\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n+ @property\n+ def secret(self) -> str | None:\n+ return self._secret\n-\n-\n-\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n+++ a/mlflow/server/handlers.py\n--- b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n+++ a/mlflow/store/artifact/azure_blob_artifact_repo.py\n--- b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+from mlflow.utils.credentials import get_default_host_creds\n-\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ augmented_raise_for_status,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(self, store_uri=None, tracking_uri=None):\n+ \"\"\"\n+ Empty constructor. This is deliberately not marked as abstract, else every derived class\n+ would be forced to create one.\n \n+ Args:\n+ store_uri: The model registry store URI.\n+ tracking_uri: URI of the current MLflow tracking server, used to perform operations\n+ like fetching source run metadata or downloading source run artifacts\n+ to support subsequently uploading them to the model registry storage\n+ location.\n+ \"\"\"\n+ # Create a thread lock to ensure thread safety when linking prompts to other entities,\n+ # since the default linking implementation reads and appends entity tags, which\n+ # is prone to concurrent modification issues\n+ self._prompt_link_lock = threading.RLock()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n+++ a/mlflow/store/model_registry/base_rest_store.py\n--- b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n+from mlflow.utils.rest_utils import (\n+ call_endpoint,\n+ call_endpoints,\n+)\n-from mlflow.utils.rest_utils import call_endpoints\n-\n-\n-\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n+++ a/mlflow/tracking/_model_registry/client.py\n--- b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ @record_usage_event(CreateWebhookEvent)\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Unique name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of event types that trigger this webhook.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE).\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ return self.store.create_webhook(name, url, events, description, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self.store.get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self.store.list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of event types.\n+ secret: New webhook secret.\n+ status: New webhook status.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self.store.delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n+ return self.store.test_webhook(webhook_id, event)\n- return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n+++ a/mlflow/tracking/_model_registry/registry.py\n--- b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n+ from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n-\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n+++ a/mlflow/tracking/_model_registry/utils.py\n--- b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n+def _get_registry_uri_from_spark_session():\n+ session = _get_active_spark_session()\n+ if session is None:\n+ return None\n \n+ if is_in_databricks_serverless_runtime():\n+ # Connected to Serverless\n+ return \"databricks-uc\"\n \n+ from pyspark.sql.utils import AnalysisException\n \n+ try:\n+ return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n+ except AnalysisException:\n+ # In serverless clusters, session.conf.get() is unsupported\n+ # and raises an AnalysisException. We may encounter this case\n+ # when DBConnect is used to connect to a serverless cluster,\n+ # in which case the prior `is_in_databricks_serverless_runtime()`\n+ # check will have returned false (as of 2025-06-07, it checks\n+ # an environment variable that isn't set by DBConnect)\n+ return None\n \n \n+def _get_registry_uri_from_context():\n+ if _registry_uri is not None:\n+ return _registry_uri\n+ elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n+ return uri\n+ return _registry_uri\n \n \n+def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n+ \"\"\"\n+ Get the default registry URI for a given tracking URI.\n \n+ If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n+ Otherwise, returns the tracking URI itself.\n \n+ Args:\n+ tracking_uri: The tracking URI to get the default registry URI for\n \n+ Returns:\n+ The default registry URI\n+ \"\"\"\n+ if tracking_uri is not None and is_databricks_uri(tracking_uri):\n+ # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n+ # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n+ # model registry offering on Databricks\n+ if tracking_uri == \"databricks\":\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+ else:\n+ # Extract profile from tracking URI and construct databricks-uc URI\n+ profile, key_prefix = get_db_info_from_uri(tracking_uri)\n+ if profile:\n+ # Reconstruct the profile string including key_prefix if present\n+ profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n+ return construct_db_uc_uri_from_profile(profile_string)\n+ else:\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+\n+ # For non-databricks tracking URIs, use the tracking URI as the registry URI\n+ return tracking_uri\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n+def _resolve_registry_uri(\n+ registry_uri: str | None = None, tracking_uri: str | None = None\n+) -> str | None:\n+ \"\"\"\n+ Resolve the registry URI following the same logic as get_registry_uri().\n+ \"\"\"\n+ return (\n+ registry_uri\n+ or _get_registry_uri_from_context()\n+ or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n+def _get_rest_store(store_uri, **_):\n+ return RestStore(partial(get_default_host_creds, store_uri))\n-\n-\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n+def _get_store_registry():\n+ global _model_registry_store_registry\n+ from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n+ from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n+ if _model_registry_store_registry is not None:\n+ return _model_registry_store_registry\n \n+ _model_registry_store_registry = ModelRegistryStoreRegistry()\n+ _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n+ # Register a placeholder function that raises if users pass a registry URI with scheme\n+ # \"databricks-uc\"\n+ _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n+ _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n+ for scheme in [\"http\", \"https\"]:\n+ _model_registry_store_registry.register(scheme, _get_rest_store)\n \n+ for scheme in DATABASE_ENGINES:\n+ _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n+ for scheme in [\"\", \"file\"]:\n+ _model_registry_store_registry.register(scheme, _get_file_store)\n \n+ _model_registry_store_registry.register_entrypoints()\n+ return _model_registry_store_registry\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n+++ a/mlflow/tracking/_tracking_service/utils.py\n--- b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n+++ a/mlflow/tracking/client.py\n--- b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n+ @property\n+ def tracking_uri(self):\n+ return self._tracking_client.tracking_uri\n \n+ def _get_registry_client(self):\n+ \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n+ Raises:\n+ MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n+ example, when the registry URI refers to an unsupported store type (e.g., the\n+ FileStore).\n \n+ Returns:\n+ A ModelRegistryClient instance.\n+ \"\"\"\n+ # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n+ # an instance variable on this `MlflowClient` instance. Because the instance variable\n+ # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n+ # function is used to safely fetch the variable (if it is defined) or a NoneType\n+ # (if it is not defined)\n+ registry_client_attr = \"_registry_client_lazy\"\n+ registry_client = getattr(self, registry_client_attr, None)\n+ if registry_client is None:\n+ try:\n+ registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n+ # Define an instance variable on this `MlflowClient` instance to reference the\n+ # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n+ # that the variable name is consistent with the variable name specified in the\n+ # preceding call to `getattr()`\n+ setattr(self, registry_client_attr, registry_client)\n+ except UnsupportedModelRegistryStoreURIException as exc:\n+ raise MlflowException(\n+ \"Model Registry features are not supported by the store with URI:\"\n+ f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n+ f\" {exc.supported_uri_schemes}.\",\n+ FEATURE_DISABLED,\n+ )\n+ return registry_client\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEventStr | WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of events that trigger this webhook. Can be strings or\n+ `WebhookEvent` objects.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self._get_registry_client().get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self._get_registry_client().list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEventStr | WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of events. Can be strings or `WebhookEvent` objects.\n+ secret: New webhook secret.\n+ status: New webhook status. Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ if events is not None:\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().update_webhook(\n+ webhook_id=webhook_id,\n+ name=name,\n+ description=description,\n+ url=url,\n+ events=events,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self._get_registry_client().delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n- return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,7 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,36 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n+++ a/mlflow/utils/credentials.py\n--- b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n+def get_default_host_creds(store_uri):\n+ creds = read_mlflow_creds()\n+ return MlflowHostCreds(\n+ host=store_uri,\n+ username=creds.username,\n+ password=creds.password,\n+ token=MLFLOW_TRACKING_TOKEN.get(),\n+ aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n+ auth=MLFLOW_TRACKING_AUTH.get(),\n+ ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n+ client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n+ server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n+def parse_dict(js_dict, message):\n+ \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n+ _stringify_all_experiment_ids(js_dict)\n+ ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n-\n-\n-\n-\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n+++ a/mlflow/utils/request_utils.py\n--- b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n+@lru_cache(maxsize=64)\n+def _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ # To create a new Session object for each process, we use the process id as the cache key.\n+ # This is to avoid sharing the same Session object across processes, which can lead to issues\n+ # such as https://stackoverflow.com/q/3724900.\n+ _pid,\n+ respect_retry_after_header=True,\n+):\n+ \"\"\"\n+ This function should not be called directly. Instead, use `_get_request_session` below.\n+ \"\"\"\n \n+ retry_kwargs = {\n+ \"total\": max_retries,\n+ \"connect\": max_retries,\n+ \"read\": max_retries,\n+ \"redirect\": max_retries,\n+ \"status\": max_retries,\n+ \"status_forcelist\": retry_codes,\n+ \"backoff_factor\": backoff_factor,\n+ \"backoff_jitter\": backoff_jitter,\n+ \"raise_on_status\": raise_on_status,\n+ \"respect_retry_after_header\": respect_retry_after_header,\n+ }\n+ urllib3_version = Version(urllib3.__version__)\n+ if urllib3_version >= Version(\"1.26.0\"):\n+ retry_kwargs[\"allowed_methods\"] = None\n+ else:\n+ retry_kwargs[\"method_whitelist\"] = None\n+\n+ if urllib3_version < Version(\"2.0\"):\n+ retry = JitteredRetry(**retry_kwargs)\n+ else:\n+ retry = Retry(**retry_kwargs)\n+ from mlflow.environment_variables import (\n+ MLFLOW_HTTP_POOL_CONNECTIONS,\n+ MLFLOW_HTTP_POOL_MAXSIZE,\n+ )\n+\n+ adapter = HTTPAdapter(\n+ pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n+ pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n+ max_retries=retry,\n+ )\n+ session = requests.Session()\n+ session.mount(\"https://\", adapter)\n+ session.mount(\"http://\", adapter)\n+ return session\n+\n+\n+def _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+):\n+ \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n+ Args:\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n+ Returns:\n+ requests.Session object.\n \n+ \"\"\"\n+ return _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ _pid=os.getpid(),\n+ respect_retry_after_header=respect_retry_after_header,\n+ )\n \n \n+def _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status=True,\n+ allow_redirects=None,\n+ respect_retry_after_header=True,\n+ **kwargs,\n+):\n+ \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n+ Args:\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ url: The target URL address for the HTTP request.\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ session = _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+ )\n \n+ # the environment variable is hardcoded here to avoid importing mlflow.\n+ # however, documentation is available in environment_variables.py\n+ env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n+ allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n+ return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n+from mlflow.utils.proto_json_utils import parse_dict\n+from mlflow.utils.request_utils import (\n+ _TRANSIENT_FAILURE_RESPONSE_CODES,\n+ _get_http_response_with_retries,\n+ augmented_raise_for_status, # noqa: F401\n+ cloud_storage_http_request, # noqa: F401\n+)\n+from mlflow.utils.string_utils import strip_suffix\n-\n-from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n-\n-\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n \n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n \n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n+\n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n \n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n+\n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n+\n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n+\n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n+\n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n+\n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n+\n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n+def _can_parse_as_json_object(string):\n+ try:\n+ return isinstance(json.loads(string), dict)\n+ except Exception:\n+ return False\n-\n-\n-\n-\n-\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n+def verify_rest_response(response, endpoint):\n+ \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n+ # Handle Armeria-specific response case where response text is \"200 OK\"\n+ # v1/traces endpoint might return empty response\n+ if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n+ response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n+ return response\n+\n+ # Handle non-200 status codes\n+ if response.status_code != 200:\n+ if _can_parse_as_json_object(response.text):\n+ raise RestException(json.loads(response.text))\n+ else:\n+ base_msg = (\n+ f\"API request to endpoint {endpoint} \"\n+ f\"failed with error code {response.status_code} != 200\"\n+ )\n+ raise MlflowException(\n+ f\"{base_msg}. Response body: '{response.text}'\",\n+ error_code=get_error_code(response.status_code),\n+ )\n \n+ # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n+ # response\n+ if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n+ base_msg = (\n+ \"API request to endpoint was successful but the response body was not \"\n+ \"in a valid JSON format\"\n+ )\n+ raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n+\n+ return response\n+\n+\n+def _validate_max_retries(max_retries):\n+ max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n+\n+ if max_retry_limit < 0:\n+ raise MlflowException(\n+ message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n+ \"Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if max_retries >= max_retry_limit:\n+ raise MlflowException(\n+ message=f\"The configured max_retries value ({max_retries}) is \"\n+ f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if max_retries < 0:\n+ raise MlflowException(\n+ message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+\n+def _validate_backoff_factor(backoff_factor):\n+ max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n+\n+ if max_backoff_factor_limit < 0:\n+ raise MlflowException(\n+ message=\"The current maximum backoff factor limit is invalid \"\n+ f\"({max_backoff_factor_limit}). Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor >= max_backoff_factor_limit:\n+ raise MlflowException(\n+ message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n+ \"of the maximum allowable backoff_factor limit \"\n+ f\"({max_backoff_factor_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor < 0:\n+ raise MlflowException(\n+ message=\"The backoff_factor value must be either 0 a positive integer. \"\n+ f\"Got {backoff_factor}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n+def call_endpoint(\n+ host_creds,\n+ endpoint,\n+ method,\n+ json_body,\n+ response_proto,\n+ extra_headers=None,\n+ retry_timeout_seconds=None,\n+):\n+ # Convert json string to json dictionary, to pass to requests\n+ if json_body is not None:\n+ json_body = json.loads(json_body)\n+ call_kwargs = {\n+ \"host_creds\": host_creds,\n+ \"endpoint\": endpoint,\n+ \"method\": method,\n+ }\n+ if extra_headers is not None:\n+ call_kwargs[\"extra_headers\"] = extra_headers\n+ if retry_timeout_seconds is not None:\n+ call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n+ if method == \"GET\":\n+ call_kwargs[\"params\"] = json_body\n+ response = http_request(**call_kwargs)\n+ else:\n+ call_kwargs[\"json\"] = json_body\n+ response = http_request(**call_kwargs)\n \n+ response = verify_rest_response(response, endpoint)\n+ response_to_parse = response.text\n+ try:\n+ js_dict = json.loads(response_to_parse)\n+ except json.JSONDecodeError:\n+ if len(response_to_parse) > 50:\n+ response_to_parse = response_to_parse[:50] + \"...\"\n+ _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n+ raise\n \n+ parse_dict(js_dict=js_dict, message=response_proto)\n+ return response_proto\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n+ def __init__(\n+ self,\n+ host,\n+ username=None,\n+ password=None,\n+ token=None,\n+ aws_sigv4=False,\n+ auth=None,\n+ ignore_tls_verification=False,\n+ client_cert_path=None,\n+ server_cert_path=None,\n+ use_databricks_sdk=False,\n+ databricks_auth_profile=None,\n+ client_id=None,\n+ client_secret=None,\n+ use_secret_scope_token=False,\n+ ):\n+ if not host:\n+ raise MlflowException(\n+ message=\"host is a required parameter for MlflowHostCreds\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if ignore_tls_verification and (server_cert_path is not None):\n+ raise MlflowException(\n+ message=(\n+ \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n+ \"must not be set! This error may have occurred because the \"\n+ \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n+ \"environment variables are both set - only one of these environment \"\n+ \"variables may be set.\"\n+ ),\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ self.host = host\n+ self.username = username\n+ self.password = password\n+ self.token = token\n+ self.aws_sigv4 = aws_sigv4\n+ self.auth = auth\n+ self.ignore_tls_verification = ignore_tls_verification\n+ self.client_cert_path = client_cert_path\n+ self.server_cert_path = server_cert_path\n+ self.use_databricks_sdk = use_databricks_sdk\n+ self.databricks_auth_profile = databricks_auth_profile\n+ self.client_id = client_id\n+ self.client_secret = client_secret\n+ self.use_secret_scope_token = use_secret_scope_token\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,11 +712,12 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n+ @property\n+ def verify(self):\n+ if self.use_databricks_sdk:\n+ # Let databricks-sdk set HTTP request `verify` param.\n+ return None\n+ if self.server_cert_path is None:\n+ return not self.ignore_tls_verification\n+ else:\n+ return self.server_cert_path\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..a8b3833e7 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -6432,6 +6432,7 @@ class MlflowClient:\n None\n \"\"\"\n self._get_registry_client().delete_webhook(webhook_id)\n+ return webhook_id\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py\ndeleted file mode 100644\nindex 97fd61b00..000000000\n--- a/tests/tracking/test_client_webhooks.py\n+++ /dev/null\n@@ -1,219 +0,0 @@\n-from pathlib import Path\n-from typing import Iterator\n-\n-import pytest\n-from cryptography.fernet import Fernet\n-\n-from mlflow.entities.webhook import WebhookAction, WebhookEntity, WebhookEvent, WebhookStatus\n-from mlflow.environment_variables import MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY\n-from mlflow.exceptions import MlflowException\n-from mlflow.server import handlers\n-from mlflow.server.fastapi_app import app\n-from mlflow.server.handlers import initialize_backend_stores\n-from mlflow.tracking import MlflowClient\n-\n-from tests.helper_functions import get_safe_port\n-from tests.tracking.integration_test_utils import ServerThread\n-\n-\n-@pytest.fixture\n-def client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Iterator[MlflowClient]:\n- \"\"\"Setup a local MLflow server with proper webhook encryption key support.\"\"\"\n- # Set up encryption key for webhooks using monkeypatch\n- encryption_key = Fernet.generate_key().decode(\"utf-8\")\n- monkeypatch.setenv(MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY.name, encryption_key)\n-\n- # Configure backend stores\n- backend_uri = f\"sqlite:///{tmp_path / 'mlflow.db'}\"\n- default_artifact_root = tmp_path.as_uri()\n-\n- # Force-reset backend stores before each test\n- handlers._tracking_store = None\n- handlers._model_registry_store = None\n- initialize_backend_stores(backend_uri, default_artifact_root=default_artifact_root)\n-\n- # Start server and return client\n- with ServerThread(app, get_safe_port()) as url:\n- yield MlflowClient(url)\n-\n-\n-def test_create_webhook(client: MlflowClient):\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n- assert webhook.events == [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n-\n- webhook = client.get_webhook(webhook.webhook_id)\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n-\n- # With secret\n- webhook_with_secret = client.create_webhook(\n- name=\"test_webhook_with_secret\",\n- url=\"https://example.com/webhook_with_secret\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- secret=\"my_secret\",\n- )\n- assert webhook_with_secret.name == \"test_webhook_with_secret\"\n- assert webhook_with_secret.url == \"https://example.com/webhook_with_secret\"\n- assert webhook_with_secret.secret is None\n- assert webhook_with_secret.events == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)\n- ]\n-\n- # Multiple events\n- webhook_multiple_events = client.create_webhook(\n- name=\"test_webhook_multiple_events\",\n- url=\"https://example.com/webhook_multiple_events\",\n- events=[\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- ],\n- )\n- assert webhook_multiple_events.name == \"test_webhook_multiple_events\"\n- assert webhook_multiple_events.url == \"https://example.com/webhook_multiple_events\"\n- assert sorted(\n- webhook_multiple_events.events, key=lambda e: (e.entity.value, e.action.value)\n- ) == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- ]\n- assert webhook_multiple_events.secret is None\n-\n-\n-def test_get_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- created_webhook = client.create_webhook(\n- name=\"test_webhook\", url=\"https://example.com/webhook\", events=events\n- )\n- retrieved_webhook = client.get_webhook(created_webhook.webhook_id)\n- assert retrieved_webhook.webhook_id == created_webhook.webhook_id\n- assert retrieved_webhook.name == \"test_webhook\"\n- assert retrieved_webhook.url == \"https://example.com/webhook\"\n- assert retrieved_webhook.events == events\n-\n-\n-def test_get_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.get_webhook(\"nonexistent\")\n-\n-\n-def test_list_webhooks(client: MlflowClient):\n- # Create more webhooks than max_results\n- created_webhooks = []\n- for i in range(5):\n- webhook = client.create_webhook(\n- name=f\"webhook{i}\",\n- url=f\"https://example.com/{i}\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- created_webhooks.append(webhook)\n- # Test pagination with max_results=2\n- webhooks_page = client.list_webhooks(max_results=2)\n- assert len(webhooks_page) == 2\n- assert webhooks_page.token is not None\n- # Get next page\n- next_webhooks_page = client.list_webhooks(max_results=2, page_token=webhooks_page.token)\n- assert len(next_webhooks_page) == 2\n- assert next_webhooks_page.token is not None\n- # Verify we don't get duplicates\n- first_page_ids = {w.webhook_id for w in webhooks_page}\n- second_page_ids = {w.webhook_id for w in next_webhooks_page}\n- assert first_page_ids.isdisjoint(second_page_ids)\n-\n-\n-def test_update_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\", url=\"https://example.com/original\", events=events\n- )\n- # Update webhook\n- new_events = [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.REGISTERED_MODEL, WebhookAction.CREATED),\n- ]\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- url=\"https://example.com/updated\",\n- events=new_events,\n- description=\"Updated description\",\n- secret=\"new_secret\",\n- status=WebhookStatus.DISABLED,\n- )\n- assert updated_webhook.webhook_id == webhook.webhook_id\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/updated\"\n- assert updated_webhook.events == new_events\n- assert updated_webhook.description == \"Updated description\"\n- assert updated_webhook.status == WebhookStatus.DISABLED\n- assert updated_webhook.last_updated_timestamp > webhook.last_updated_timestamp\n-\n-\n-def test_update_webhook_partial(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\",\n- url=\"https://example.com/original\",\n- events=events,\n- description=\"Original description\",\n- )\n- # Update only the name\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- )\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/original\"\n- assert updated_webhook.events == events\n- assert updated_webhook.description == \"Original description\"\n-\n-\n-def test_update_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.update_webhook(webhook_id=\"nonexistent\", name=\"new_name\")\n-\n-\n-@pytest.mark.parametrize(\n- (\"invalid_url\", \"expected_match\"),\n- [\n- (\" \", r\"Webhook URL cannot be empty or just whitespace\"),\n- (\"ftp://example.com\", r\"Invalid webhook URL scheme\"),\n- (\"http://[invalid\", r\"Invalid webhook URL\"),\n- ],\n-)\n-def test_update_webhook_invalid_urls(client: MlflowClient, invalid_url: str, expected_match: str):\n- # Create a valid webhook first\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- with pytest.raises(MlflowException, match=expected_match):\n- client.update_webhook(webhook_id=webhook.webhook_id, url=invalid_url)\n-\n-\n-def test_delete_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=events,\n- )\n- client.delete_webhook(webhook.webhook_id)\n- with pytest.raises(MlflowException, match=r\"Webhook with ID .* not found\"):\n- client.get_webhook(webhook.webhook_id)\n- webhooks_page = client.list_webhooks()\n- webhook_ids = {w.webhook_id for w in webhooks_page}\n- assert webhook.webhook_id not in webhook_ids\n-\n-\n-def test_delete_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.delete_webhook(\"nonexistent\")\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S14\",\n \"shared_functions\": 2,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":62,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S14\",\n \"text\": \"Raises: MlflowException: If the webhook with the specified ID does not exist, if there are permission issues, or if the deletion operation fails for any other reason.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracking/client.py\",\n \"lines\": \"6437\",\n \"function\": \"MlflowClient.delete_webhook\"\n },\n {\n \"file\": \"mlflow/tracking/_model_registry/client.py\",\n \"lines\": \"895\",\n \"function\": \"ModelRegistryClient.delete_webhook\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n--- a/mlflow/entities/webhook.py\n+++ b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n- def __eq__(self, other: object) -> bool:\n- if not isinstance(other, WebhookEvent):\n- return False\n- return self.entity == other.entity and self.action == other.action\n+\n+\n+\n+\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n- @property\n- def secret(self) -> str | None:\n- return self._secret\n+\n+\n+\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n--- a/mlflow/server/handlers.py\n+++ b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n--- a/mlflow/store/artifact/azure_blob_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-from mlflow.utils.credentials import get_default_host_creds\n+\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- augmented_raise_for_status,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _UC_OSS_REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n+\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n- def __init__(self, store_uri=None, tracking_uri=None):\n- \"\"\"\n- Empty constructor. This is deliberately not marked as abstract, else every derived class\n- would be forced to create one.\n \n- Args:\n- store_uri: The model registry store URI.\n- tracking_uri: URI of the current MLflow tracking server, used to perform operations\n- like fetching source run metadata or downloading source run artifacts\n- to support subsequently uploading them to the model registry storage\n- location.\n- \"\"\"\n- # Create a thread lock to ensure thread safety when linking prompts to other entities,\n- # since the default linking implementation reads and appends entity tags, which\n- # is prone to concurrent modification issues\n- self._prompt_link_lock = threading.RLock()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n--- a/mlflow/store/model_registry/base_rest_store.py\n+++ b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n-from mlflow.utils.rest_utils import (\n- call_endpoint,\n- call_endpoints,\n-)\n+from mlflow.utils.rest_utils import call_endpoints\n+\n+\n+\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n+\n+\n+\n+\n+\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- @record_usage_event(CreateWebhookEvent)\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Unique name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of event types that trigger this webhook.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE).\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- return self.store.create_webhook(name, url, events, description, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self.store.get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self.store.list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of event types.\n- secret: New webhook secret.\n- status: New webhook status.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self.store.delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n- return self.store.test_webhook(webhook_id, event)\n+ return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n--- a/mlflow/tracking/_model_registry/registry.py\n+++ b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n- from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n+\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n--- a/mlflow/tracking/_model_registry/utils.py\n+++ b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n-def _get_registry_uri_from_spark_session():\n- session = _get_active_spark_session()\n- if session is None:\n- return None\n \n- if is_in_databricks_serverless_runtime():\n- # Connected to Serverless\n- return \"databricks-uc\"\n \n- from pyspark.sql.utils import AnalysisException\n \n- try:\n- return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n- except AnalysisException:\n- # In serverless clusters, session.conf.get() is unsupported\n- # and raises an AnalysisException. We may encounter this case\n- # when DBConnect is used to connect to a serverless cluster,\n- # in which case the prior `is_in_databricks_serverless_runtime()`\n- # check will have returned false (as of 2025-06-07, it checks\n- # an environment variable that isn't set by DBConnect)\n- return None\n \n \n-def _get_registry_uri_from_context():\n- if _registry_uri is not None:\n- return _registry_uri\n- elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n- return uri\n- return _registry_uri\n \n \n-def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n- \"\"\"\n- Get the default registry URI for a given tracking URI.\n \n- If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n- Otherwise, returns the tracking URI itself.\n \n- Args:\n- tracking_uri: The tracking URI to get the default registry URI for\n \n- Returns:\n- The default registry URI\n- \"\"\"\n- if tracking_uri is not None and is_databricks_uri(tracking_uri):\n- # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n- # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n- # model registry offering on Databricks\n- if tracking_uri == \"databricks\":\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n- else:\n- # Extract profile from tracking URI and construct databricks-uc URI\n- profile, key_prefix = get_db_info_from_uri(tracking_uri)\n- if profile:\n- # Reconstruct the profile string including key_prefix if present\n- profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n- return construct_db_uc_uri_from_profile(profile_string)\n- else:\n- return _DATABRICKS_UNITY_CATALOG_SCHEME\n-\n- # For non-databricks tracking URIs, use the tracking URI as the registry URI\n- return tracking_uri\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n-def _resolve_registry_uri(\n- registry_uri: str | None = None, tracking_uri: str | None = None\n-) -> str | None:\n- \"\"\"\n- Resolve the registry URI following the same logic as get_registry_uri().\n- \"\"\"\n- return (\n- registry_uri\n- or _get_registry_uri_from_context()\n- or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n-def _get_rest_store(store_uri, **_):\n- return RestStore(partial(get_default_host_creds, store_uri))\n+\n+\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n-def _get_store_registry():\n- global _model_registry_store_registry\n- from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n- from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n- if _model_registry_store_registry is not None:\n- return _model_registry_store_registry\n \n- _model_registry_store_registry = ModelRegistryStoreRegistry()\n- _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n- # Register a placeholder function that raises if users pass a registry URI with scheme\n- # \"databricks-uc\"\n- _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n- _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n- for scheme in [\"http\", \"https\"]:\n- _model_registry_store_registry.register(scheme, _get_rest_store)\n \n- for scheme in DATABASE_ENGINES:\n- _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n- for scheme in [\"\", \"file\"]:\n- _model_registry_store_registry.register(scheme, _get_file_store)\n \n- _model_registry_store_registry.register_entrypoints()\n- return _model_registry_store_registry\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n--- a/mlflow/tracking/_tracking_service/utils.py\n+++ b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n-from mlflow.utils.credentials import get_default_host_creds\n+\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n--- a/mlflow/tracking/client.py\n+++ b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n- @property\n- def tracking_uri(self):\n- return self._tracking_client.tracking_uri\n \n- def _get_registry_client(self):\n- \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n- Raises:\n- MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n- example, when the registry URI refers to an unsupported store type (e.g., the\n- FileStore).\n \n- Returns:\n- A ModelRegistryClient instance.\n- \"\"\"\n- # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n- # an instance variable on this `MlflowClient` instance. Because the instance variable\n- # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n- # function is used to safely fetch the variable (if it is defined) or a NoneType\n- # (if it is not defined)\n- registry_client_attr = \"_registry_client_lazy\"\n- registry_client = getattr(self, registry_client_attr, None)\n- if registry_client is None:\n- try:\n- registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n- # Define an instance variable on this `MlflowClient` instance to reference the\n- # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n- # that the variable name is consistent with the variable name specified in the\n- # preceding call to `getattr()`\n- setattr(self, registry_client_attr, registry_client)\n- except UnsupportedModelRegistryStoreURIException as exc:\n- raise MlflowException(\n- \"Model Registry features are not supported by the store with URI:\"\n- f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n- f\" {exc.supported_uri_schemes}.\",\n- FEATURE_DISABLED,\n- )\n- return registry_client\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n- @experimental(version=\"3.3.0\")\n- def create_webhook(\n- self,\n- name: str,\n- url: str,\n- events: list[WebhookEventStr | WebhookEvent],\n- description: str | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Create a new webhook.\n \n- Args:\n- name: Name for the webhook.\n- url: Webhook endpoint URL.\n- events: List of events that trigger this webhook. Can be strings or\n- `WebhookEvent` objects.\n- description: Optional description of the webhook.\n- secret: Optional secret for HMAC signature verification.\n- status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n- \"\"\"\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().create_webhook(\n- name=name,\n- url=url,\n- events=events,\n- description=description,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def get_webhook(self, webhook_id: str) -> Webhook:\n- \"\"\"\n- Get webhook instance by ID.\n \n- Args:\n- webhook_id: Webhook ID.\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object.\n- \"\"\"\n- return self._get_registry_client().get_webhook(webhook_id)\n \n- @experimental(version=\"3.3.0\")\n- def list_webhooks(\n- self,\n- max_results: int | None = None,\n- page_token: str | None = None,\n- ) -> PagedList[Webhook]:\n- \"\"\"\n- List webhooks.\n \n- Args:\n- max_results: Maximum number of webhooks to return.\n- page_token: Token specifying the next page of results.\n \n- Returns:\n- A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n- \"\"\"\n- return self._get_registry_client().list_webhooks(max_results, page_token)\n \n- @experimental(version=\"3.3.0\")\n- def update_webhook(\n- self,\n- webhook_id: str,\n- name: str | None = None,\n- description: str | None = None,\n- url: str | None = None,\n- events: list[WebhookEventStr | WebhookEvent] | None = None,\n- secret: str | None = None,\n- status: str | WebhookStatus | None = None,\n- ) -> Webhook:\n- \"\"\"\n- Update an existing webhook.\n \n- Args:\n- webhook_id: Webhook ID.\n- name: New webhook name.\n- description: New webhook description.\n- url: New webhook URL.\n- events: New list of events. Can be strings or `WebhookEvent` objects.\n- secret: New webhook secret.\n- status: New webhook status. Can be string or WebhookStatus enum.\n- Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n- Returns:\n- A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n- \"\"\"\n- if events is not None:\n- events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n- if status is not None:\n- status = WebhookStatus(status) if isinstance(status, str) else status\n \n- return self._get_registry_client().update_webhook(\n- webhook_id=webhook_id,\n- name=name,\n- description=description,\n- url=url,\n- events=events,\n- secret=secret,\n- status=status,\n- )\n \n- @experimental(version=\"3.3.0\")\n- def delete_webhook(self, webhook_id: str) -> None:\n- \"\"\"\n- Delete a webhook.\n \n- Args:\n- webhook_id: Webhook ID to delete.\n \n- Returns:\n- None\n- \"\"\"\n- self._get_registry_client().delete_webhook(webhook_id)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n- return self._get_registry_client().test_webhook(webhook_id, event)\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,7 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,36 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_api_info_for_service,\n-)\n+\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n--- a/mlflow/utils/credentials.py\n+++ b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n-def get_default_host_creds(store_uri):\n- creds = read_mlflow_creds()\n- return MlflowHostCreds(\n- host=store_uri,\n- username=creds.username,\n- password=creds.password,\n- token=MLFLOW_TRACKING_TOKEN.get(),\n- aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n- auth=MLFLOW_TRACKING_AUTH.get(),\n- ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n- client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n- server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n-def parse_dict(js_dict, message):\n- \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n- _stringify_all_experiment_ids(js_dict)\n- ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n+\n+\n+\n+\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n--- a/mlflow/utils/request_utils.py\n+++ b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n-@lru_cache(maxsize=64)\n-def _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- # To create a new Session object for each process, we use the process id as the cache key.\n- # This is to avoid sharing the same Session object across processes, which can lead to issues\n- # such as https://stackoverflow.com/q/3724900.\n- _pid,\n- respect_retry_after_header=True,\n-):\n- \"\"\"\n- This function should not be called directly. Instead, use `_get_request_session` below.\n- \"\"\"\n \n- retry_kwargs = {\n- \"total\": max_retries,\n- \"connect\": max_retries,\n- \"read\": max_retries,\n- \"redirect\": max_retries,\n- \"status\": max_retries,\n- \"status_forcelist\": retry_codes,\n- \"backoff_factor\": backoff_factor,\n- \"backoff_jitter\": backoff_jitter,\n- \"raise_on_status\": raise_on_status,\n- \"respect_retry_after_header\": respect_retry_after_header,\n- }\n- urllib3_version = Version(urllib3.__version__)\n- if urllib3_version >= Version(\"1.26.0\"):\n- retry_kwargs[\"allowed_methods\"] = None\n- else:\n- retry_kwargs[\"method_whitelist\"] = None\n-\n- if urllib3_version < Version(\"2.0\"):\n- retry = JitteredRetry(**retry_kwargs)\n- else:\n- retry = Retry(**retry_kwargs)\n- from mlflow.environment_variables import (\n- MLFLOW_HTTP_POOL_CONNECTIONS,\n- MLFLOW_HTTP_POOL_MAXSIZE,\n- )\n-\n- adapter = HTTPAdapter(\n- pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n- pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n- max_retries=retry,\n- )\n- session = requests.Session()\n- session.mount(\"https://\", adapter)\n- session.mount(\"http://\", adapter)\n- return session\n-\n-\n-def _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n-):\n- \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n- Args:\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n- Returns:\n- requests.Session object.\n \n- \"\"\"\n- return _cached_get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- _pid=os.getpid(),\n- respect_retry_after_header=respect_retry_after_header,\n- )\n \n \n-def _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status=True,\n- allow_redirects=None,\n- respect_retry_after_header=True,\n- **kwargs,\n-):\n- \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n- Args:\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- url: The target URL address for the HTTP request.\n- max_retries: Maximum total number of retries.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- session = _get_request_session(\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- respect_retry_after_header,\n- )\n \n- # the environment variable is hardcoded here to avoid importing mlflow.\n- # however, documentation is available in environment_variables.py\n- env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n- allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n- return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n-from mlflow.utils.proto_json_utils import parse_dict\n-from mlflow.utils.request_utils import (\n- _TRANSIENT_FAILURE_RESPONSE_CODES,\n- _get_http_response_with_retries,\n- augmented_raise_for_status, # noqa: F401\n- cloud_storage_http_request, # noqa: F401\n-)\n-from mlflow.utils.string_utils import strip_suffix\n+\n+from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n+\n+\n+\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n \n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n \n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n-\n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n \n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n-\n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n-\n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n-\n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n-\n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n-\n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n-\n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n-def _can_parse_as_json_object(string):\n- try:\n- return isinstance(json.loads(string), dict)\n- except Exception:\n- return False\n+\n+\n+\n+\n+\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n-def verify_rest_response(response, endpoint):\n- \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n- # Handle Armeria-specific response case where response text is \"200 OK\"\n- # v1/traces endpoint might return empty response\n- if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n- response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n- return response\n-\n- # Handle non-200 status codes\n- if response.status_code != 200:\n- if _can_parse_as_json_object(response.text):\n- raise RestException(json.loads(response.text))\n- else:\n- base_msg = (\n- f\"API request to endpoint {endpoint} \"\n- f\"failed with error code {response.status_code} != 200\"\n- )\n- raise MlflowException(\n- f\"{base_msg}. Response body: '{response.text}'\",\n- error_code=get_error_code(response.status_code),\n- )\n \n- # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n- # response\n- if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n- base_msg = (\n- \"API request to endpoint was successful but the response body was not \"\n- \"in a valid JSON format\"\n- )\n- raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n-\n- return response\n-\n-\n-def _validate_max_retries(max_retries):\n- max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n-\n- if max_retry_limit < 0:\n- raise MlflowException(\n- message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n- \"Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if max_retries >= max_retry_limit:\n- raise MlflowException(\n- message=f\"The configured max_retries value ({max_retries}) is \"\n- f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if max_retries < 0:\n- raise MlflowException(\n- message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n-\n-def _validate_backoff_factor(backoff_factor):\n- max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n-\n- if max_backoff_factor_limit < 0:\n- raise MlflowException(\n- message=\"The current maximum backoff factor limit is invalid \"\n- f\"({max_backoff_factor_limit}). Cannot be negative.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor >= max_backoff_factor_limit:\n- raise MlflowException(\n- message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n- \"of the maximum allowable backoff_factor limit \"\n- f\"({max_backoff_factor_limit})\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n-\n- if backoff_factor < 0:\n- raise MlflowException(\n- message=\"The backoff_factor value must be either 0 a positive integer. \"\n- f\"Got {backoff_factor}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n-def call_endpoint(\n- host_creds,\n- endpoint,\n- method,\n- json_body,\n- response_proto,\n- extra_headers=None,\n- retry_timeout_seconds=None,\n-):\n- # Convert json string to json dictionary, to pass to requests\n- if json_body is not None:\n- json_body = json.loads(json_body)\n- call_kwargs = {\n- \"host_creds\": host_creds,\n- \"endpoint\": endpoint,\n- \"method\": method,\n- }\n- if extra_headers is not None:\n- call_kwargs[\"extra_headers\"] = extra_headers\n- if retry_timeout_seconds is not None:\n- call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n- if method == \"GET\":\n- call_kwargs[\"params\"] = json_body\n- response = http_request(**call_kwargs)\n- else:\n- call_kwargs[\"json\"] = json_body\n- response = http_request(**call_kwargs)\n \n- response = verify_rest_response(response, endpoint)\n- response_to_parse = response.text\n- try:\n- js_dict = json.loads(response_to_parse)\n- except json.JSONDecodeError:\n- if len(response_to_parse) > 50:\n- response_to_parse = response_to_parse[:50] + \"...\"\n- _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n- raise\n \n- parse_dict(js_dict=js_dict, message=response_proto)\n- return response_proto\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n- def __init__(\n- self,\n- host,\n- username=None,\n- password=None,\n- token=None,\n- aws_sigv4=False,\n- auth=None,\n- ignore_tls_verification=False,\n- client_cert_path=None,\n- server_cert_path=None,\n- use_databricks_sdk=False,\n- databricks_auth_profile=None,\n- client_id=None,\n- client_secret=None,\n- use_secret_scope_token=False,\n- ):\n- if not host:\n- raise MlflowException(\n- message=\"host is a required parameter for MlflowHostCreds\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- if ignore_tls_verification and (server_cert_path is not None):\n- raise MlflowException(\n- message=(\n- \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n- \"must not be set! This error may have occurred because the \"\n- \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n- \"environment variables are both set - only one of these environment \"\n- \"variables may be set.\"\n- ),\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- self.host = host\n- self.username = username\n- self.password = password\n- self.token = token\n- self.aws_sigv4 = aws_sigv4\n- self.auth = auth\n- self.ignore_tls_verification = ignore_tls_verification\n- self.client_cert_path = client_cert_path\n- self.server_cert_path = server_cert_path\n- self.use_databricks_sdk = use_databricks_sdk\n- self.databricks_auth_profile = databricks_auth_profile\n- self.client_id = client_id\n- self.client_secret = client_secret\n- self.use_secret_scope_token = use_secret_scope_token\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,12 +712,11 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n- @property\n- def verify(self):\n- if self.use_databricks_sdk:\n- # Let databricks-sdk set HTTP request `verify` param.\n- return None\n- if self.server_cert_path is None:\n- return not self.ignore_tls_verification\n- else:\n- return self.server_cert_path\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..bd1483000 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.string_utils import _backtick_quote\n \n DATABRICKS_HIVE_METASTORE_NAME = \"hive_metastore\"\ndiff --git a/mlflow/entities/webhook.py b/mlflow/entities/webhook.py\nindex d3db3feb9..dfffba3bd 100644\n+++ a/mlflow/entities/webhook.py\n--- b/mlflow/entities/webhook.py\n@@ -235,10 +235,10 @@ class WebhookEvent:\n def __str__(self) -> str:\n return f\"{self.entity.value}.{self.action.value}\"\n \n+ def __eq__(self, other: object) -> bool:\n+ if not isinstance(other, WebhookEvent):\n+ return False\n+ return self.entity == other.entity and self.action == other.action\n-\n-\n-\n-\n \n def __hash__(self) -> int:\n return hash((self.entity, self.action))\n@@ -315,9 +315,9 @@ class Webhook:\n def status(self) -> WebhookStatus:\n return self._status\n \n+ @property\n+ def secret(self) -> str | None:\n+ return self._secret\n-\n-\n-\n \n @property\n def creation_timestamp(self) -> int:\n@@ -434,4 +434,4 @@ class WebhookTestResult:\n f\"response_body={self.response_body!r}, \"\n f\"error_message={self.error_message!r}\"\n f\")\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/server/handlers.py b/mlflow/server/handlers.py\nindex 68db503d2..6b53bbb69 100644\n+++ a/mlflow/server/handlers.py\n--- b/mlflow/server/handlers.py\n@@ -202,7 +202,7 @@ from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import local_file_uri_to_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n from mlflow.utils.promptlab_utils import _create_promptlab_run_impl\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-\n from mlflow.utils.string_utils import is_string_type\n from mlflow.utils.uri import is_local_uri, validate_path_is_safe, validate_query_string\n from mlflow.utils.validation import (\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..535ae092b 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,15 +157,15 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..81bb67d16 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,13 +39,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _UC_OSS_REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_file_uri, is_fuse_or_uc_volumes_uri\n \n _METHOD_TO_INFO = extract_api_info_for_service(UnityCatalogService, _UC_OSS_REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/azure_blob_artifact_repo.py b/mlflow/store/artifact/azure_blob_artifact_repo.py\nindex b1e7f6277..155d16121 100644\n+++ a/mlflow/store/artifact/azure_blob_artifact_repo.py\n--- b/mlflow/store/artifact/azure_blob_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.entities.multipart_upload import (\n from mlflow.environment_variables import MLFLOW_ARTIFACT_UPLOAD_DOWNLOAD_TIMEOUT\n from mlflow.exceptions import MlflowException\n from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+from mlflow.utils.credentials import get_default_host_creds\n-\n \n \n def encode_base64(data: str | bytes) -> str:\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..ad669eec1 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,14 +62,14 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ augmented_raise_for_status,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, augmented_raise_for_status, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n extract_and_normalize_path,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..ee81828b1 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -23,10 +23,10 @@ from mlflow.store.artifact.artifact_repo import (\n verify_artifact_path,\n )\n from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _compute_num_chunks\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..cb213bbca 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -18,13 +18,13 @@ from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n FILESYSTEM_METHOD_TO_INFO = extract_api_info_for_service(FilesystemService, _REST_API_PATH_PREFIX)\n DIRECTORIES_ENDPOINT = \"/api/2.0/fs/directories\"\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..bb6b0d2f9 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,12 +27,12 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..f8f7782da 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,13 +24,13 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _UC_OSS_REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _UC_OSS_REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n-\n from mlflow.utils.uri import (\n _OSS_UNITY_CATALOG_SCHEME,\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1f137ec3b 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -49,22 +49,22 @@ class AbstractStore:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(self, store_uri=None, tracking_uri=None):\n+ \"\"\"\n+ Empty constructor. This is deliberately not marked as abstract, else every derived class\n+ would be forced to create one.\n \n+ Args:\n+ store_uri: The model registry store URI.\n+ tracking_uri: URI of the current MLflow tracking server, used to perform operations\n+ like fetching source run metadata or downloading source run artifacts\n+ to support subsequently uploading them to the model registry storage\n+ location.\n+ \"\"\"\n+ # Create a thread lock to ensure thread safety when linking prompts to other entities,\n+ # since the default linking implementation reads and appends entity tags, which\n+ # is prone to concurrent modification issues\n+ self._prompt_link_lock = threading.RLock()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __getstate__(self):\n \"\"\"Support for pickle serialization by excluding the non-picklable RLock.\"\"\"\n@@ -1179,4 +1179,4 @@ class AbstractStore:\n Returns:\n WebhookTestResult indicating success/failure and response details\n \"\"\"\n+ raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n- raise NotImplementedError(f\"{self.__class__.__name__} does not support test_webhook\")\n\\ No newline at end of file\ndiff --git a/mlflow/store/model_registry/base_rest_store.py b/mlflow/store/model_registry/base_rest_store.py\nindex c0fbb103e..a99ad4474 100644\n+++ a/mlflow/store/model_registry/base_rest_store.py\n--- b/mlflow/store/model_registry/base_rest_store.py\n@@ -1,10 +1,10 @@\n from abc import ABCMeta, abstractmethod\n \n from mlflow.store.model_registry.abstract_store import AbstractStore\n+from mlflow.utils.rest_utils import (\n+ call_endpoint,\n+ call_endpoints,\n+)\n-from mlflow.utils.rest_utils import call_endpoints\n-\n-\n-\n \n \n class BaseRestStore(AbstractStore):\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..3255e34c9 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,13 +37,13 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_all_api_info_for_service, extract_api_info_for_service\n-\n-\n-\n-\n-\n \n _METHOD_TO_INFO = extract_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\n _METHOD_TO_ALL_INFO = extract_all_api_info_for_service(ModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..169fb84ee 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,17 +52,17 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..e6d33ad53 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,20 +120,20 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..64e3b49c6 100644\n+++ a/mlflow/tracking/_model_registry/client.py\n--- b/mlflow/tracking/_model_registry/client.py\n@@ -795,104 +795,104 @@ class ModelRegistryClient:\n self.store.delete_prompt_version_tag(name, version, key)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ @record_usage_event(CreateWebhookEvent)\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Unique name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of event types that trigger this webhook.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE).\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ return self.store.create_webhook(name, url, events, description, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self.store.get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self.store.list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of event types.\n+ secret: New webhook secret.\n+ status: New webhook status.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ return self.store.update_webhook(webhook_id, name, description, url, events, secret, status)\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self.store.delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -914,4 +914,4 @@ class ModelRegistryClient:\n # Convert string to WebhookEvent if needed\n if isinstance(event, str):\n event = WebhookEvent.from_str(event)\n+ return self.store.test_webhook(webhook_id, event)\n- return self.store.test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_model_registry/registry.py b/mlflow/tracking/_model_registry/registry.py\nindex 049602285..6d5b2babd 100644\n+++ a/mlflow/tracking/_model_registry/registry.py\n--- b/mlflow/tracking/_model_registry/registry.py\n@@ -41,7 +41,7 @@ class ModelRegistryStoreRegistry(StoreRegistry):\n store URI requirements.\n \n \"\"\"\n+ from mlflow.tracking._model_registry.utils import _resolve_registry_uri\n-\n from mlflow.tracking._tracking_service.utils import _resolve_tracking_uri\n \n resolved_store_uri = _resolve_registry_uri(store_uri)\ndiff --git a/mlflow/tracking/_model_registry/utils.py b/mlflow/tracking/_model_registry/utils.py\nindex 210138891..4c4e79cc2 100644\n+++ a/mlflow/tracking/_model_registry/utils.py\n--- b/mlflow/tracking/_model_registry/utils.py\n@@ -12,7 +12,7 @@ from mlflow.tracking._tracking_service.utils import (\n _resolve_tracking_uri,\n )\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import (\n is_in_databricks_serverless_runtime,\n warn_on_deprecated_cross_workspace_registry_uri,\n@@ -86,68 +86,68 @@ def set_registry_uri(uri: str) -> None:\n MLFLOW_REGISTRY_URI.set(_registry_uri)\n \n \n+def _get_registry_uri_from_spark_session():\n+ session = _get_active_spark_session()\n+ if session is None:\n+ return None\n \n+ if is_in_databricks_serverless_runtime():\n+ # Connected to Serverless\n+ return \"databricks-uc\"\n \n+ from pyspark.sql.utils import AnalysisException\n \n+ try:\n+ return session.conf.get(\"spark.mlflow.modelRegistryUri\", None)\n+ except AnalysisException:\n+ # In serverless clusters, session.conf.get() is unsupported\n+ # and raises an AnalysisException. We may encounter this case\n+ # when DBConnect is used to connect to a serverless cluster,\n+ # in which case the prior `is_in_databricks_serverless_runtime()`\n+ # check will have returned false (as of 2025-06-07, it checks\n+ # an environment variable that isn't set by DBConnect)\n+ return None\n \n \n+def _get_registry_uri_from_context():\n+ if _registry_uri is not None:\n+ return _registry_uri\n+ elif (uri := MLFLOW_REGISTRY_URI.get()) or (uri := _get_registry_uri_from_spark_session()):\n+ return uri\n+ return _registry_uri\n \n \n+def _get_default_registry_uri_for_tracking_uri(tracking_uri: str | None) -> str | None:\n+ \"\"\"\n+ Get the default registry URI for a given tracking URI.\n \n+ If the tracking URI starts with \"databricks\", returns \"databricks-uc\" with profile if present.\n+ Otherwise, returns the tracking URI itself.\n \n+ Args:\n+ tracking_uri: The tracking URI to get the default registry URI for\n \n+ Returns:\n+ The default registry URI\n+ \"\"\"\n+ if tracking_uri is not None and is_databricks_uri(tracking_uri):\n+ # If the tracking URI is \"databricks\", we impute the registry URI as \"databricks-uc\"\n+ # corresponding to Databricks Unity Catalog Model Registry, which is the recommended\n+ # model registry offering on Databricks\n+ if tracking_uri == \"databricks\":\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+ else:\n+ # Extract profile from tracking URI and construct databricks-uc URI\n+ profile, key_prefix = get_db_info_from_uri(tracking_uri)\n+ if profile:\n+ # Reconstruct the profile string including key_prefix if present\n+ profile_string = f\"{profile}:{key_prefix}\" if key_prefix else profile\n+ return construct_db_uc_uri_from_profile(profile_string)\n+ else:\n+ return _DATABRICKS_UNITY_CATALOG_SCHEME\n+\n+ # For non-databricks tracking URIs, use the tracking URI as the registry URI\n+ return tracking_uri\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def get_registry_uri() -> str:\n@@ -178,17 +178,17 @@ def get_registry_uri() -> str:\n return _resolve_registry_uri()\n \n \n+def _resolve_registry_uri(\n+ registry_uri: str | None = None, tracking_uri: str | None = None\n+) -> str | None:\n+ \"\"\"\n+ Resolve the registry URI following the same logic as get_registry_uri().\n+ \"\"\"\n+ return (\n+ registry_uri\n+ or _get_registry_uri_from_context()\n+ or _get_default_registry_uri_for_tracking_uri(_resolve_tracking_uri(tracking_uri))\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_sqlalchemy_store(store_uri):\n@@ -197,8 +197,8 @@ def _get_sqlalchemy_store(store_uri):\n return SqlAlchemyStore(store_uri)\n \n \n+def _get_rest_store(store_uri, **_):\n+ return RestStore(partial(get_default_host_creds, store_uri))\n-\n-\n \n \n def _get_databricks_rest_store(store_uri, tracking_uri, **_):\n@@ -215,33 +215,33 @@ def _get_file_store(store_uri, **_):\n return FileStore(store_uri)\n \n \n+def _get_store_registry():\n+ global _model_registry_store_registry\n+ from mlflow.store._unity_catalog.registry.rest_store import UcModelRegistryStore\n+ from mlflow.store._unity_catalog.registry.uc_oss_rest_store import UnityCatalogOssStore\n \n+ if _model_registry_store_registry is not None:\n+ return _model_registry_store_registry\n \n+ _model_registry_store_registry = ModelRegistryStoreRegistry()\n+ _model_registry_store_registry.register(\"databricks\", _get_databricks_rest_store)\n+ # Register a placeholder function that raises if users pass a registry URI with scheme\n+ # \"databricks-uc\"\n+ _model_registry_store_registry.register(_DATABRICKS_UNITY_CATALOG_SCHEME, UcModelRegistryStore)\n+ _model_registry_store_registry.register(_OSS_UNITY_CATALOG_SCHEME, UnityCatalogOssStore)\n \n+ for scheme in [\"http\", \"https\"]:\n+ _model_registry_store_registry.register(scheme, _get_rest_store)\n \n+ for scheme in DATABASE_ENGINES:\n+ _model_registry_store_registry.register(scheme, _get_sqlalchemy_store)\n \n+ for scheme in [\"\", \"file\"]:\n+ _model_registry_store_registry.register(scheme, _get_file_store)\n \n+ _model_registry_store_registry.register_entrypoints()\n+ return _model_registry_store_registry\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_store(store_uri=None, tracking_uri=None):\n+ return _get_store_registry().get_store(store_uri, tracking_uri)\n- return _get_store_registry().get_store(store_uri, tracking_uri)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/_tracking_service/utils.py b/mlflow/tracking/_tracking_service/utils.py\nindex cb7e0689f..f97b9a107 100644\n+++ a/mlflow/tracking/_tracking_service/utils.py\n--- b/mlflow/tracking/_tracking_service/utils.py\n@@ -13,7 +13,7 @@ from mlflow.store.tracking.databricks_rest_store import DatabricksTracingRestSto\n from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracing.provider import reset\n from mlflow.tracking._tracking_service.registry import TrackingStoreRegistry\n+from mlflow.utils.credentials import get_default_host_creds\n-\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import path_to_local_file_uri\n from mlflow.utils.uri import (\ndiff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py\nindex 8a34c23cd..cdb0c83ce 100644\n+++ a/mlflow/tracking/client.py\n--- b/mlflow/tracking/client.py\n@@ -224,44 +224,44 @@ class MlflowClient:\n # is assigned lazily by `MlflowClient._get_registry_client()` and should not be referenced\n # outside of the `MlflowClient._get_registry_client()` method\n \n+ @property\n+ def tracking_uri(self):\n+ return self._tracking_client.tracking_uri\n \n+ def _get_registry_client(self):\n+ \"\"\"Attempts to create a ModelRegistryClient if one does not already exist.\n \n+ Raises:\n+ MlflowException: If the ModelRegistryClient cannot be created. This may occur, for\n+ example, when the registry URI refers to an unsupported store type (e.g., the\n+ FileStore).\n \n+ Returns:\n+ A ModelRegistryClient instance.\n+ \"\"\"\n+ # Attempt to fetch a `ModelRegistryClient` that is lazily instantiated and defined as\n+ # an instance variable on this `MlflowClient` instance. Because the instance variable\n+ # is undefined until the first invocation of _get_registry_client(), the `getattr()`\n+ # function is used to safely fetch the variable (if it is defined) or a NoneType\n+ # (if it is not defined)\n+ registry_client_attr = \"_registry_client_lazy\"\n+ registry_client = getattr(self, registry_client_attr, None)\n+ if registry_client is None:\n+ try:\n+ registry_client = ModelRegistryClient(self._registry_uri, self.tracking_uri)\n+ # Define an instance variable on this `MlflowClient` instance to reference the\n+ # `ModelRegistryClient` that was just constructed. `setattr()` is used to ensure\n+ # that the variable name is consistent with the variable name specified in the\n+ # preceding call to `getattr()`\n+ setattr(self, registry_client_attr, registry_client)\n+ except UnsupportedModelRegistryStoreURIException as exc:\n+ raise MlflowException(\n+ \"Model Registry features are not supported by the store with URI:\"\n+ f\" '{self._registry_uri}'. Stores with the following URI schemes are supported:\"\n+ f\" {exc.supported_uri_schemes}.\",\n+ FEATURE_DISABLED,\n+ )\n+ return registry_client\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n # Tracking API\n \n@@ -6307,131 +6307,131 @@ class MlflowClient:\n return self._tracking_client.remove_dataset_from_experiments(dataset_id, experiment_ids)\n \n # Webhook APIs\n+ @experimental(version=\"3.3.0\")\n+ def create_webhook(\n+ self,\n+ name: str,\n+ url: str,\n+ events: list[WebhookEventStr | WebhookEvent],\n+ description: str | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Create a new webhook.\n \n+ Args:\n+ name: Name for the webhook.\n+ url: Webhook endpoint URL.\n+ events: List of events that trigger this webhook. Can be strings or\n+ `WebhookEvent` objects.\n+ description: Optional description of the webhook.\n+ secret: Optional secret for HMAC signature verification.\n+ status: Webhook status (defaults to ACTIVE). Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the created webhook.\n+ \"\"\"\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().create_webhook(\n+ name=name,\n+ url=url,\n+ events=events,\n+ description=description,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def get_webhook(self, webhook_id: str) -> Webhook:\n+ \"\"\"\n+ Get webhook instance by ID.\n \n+ Args:\n+ webhook_id: Webhook ID.\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object.\n+ \"\"\"\n+ return self._get_registry_client().get_webhook(webhook_id)\n \n+ @experimental(version=\"3.3.0\")\n+ def list_webhooks(\n+ self,\n+ max_results: int | None = None,\n+ page_token: str | None = None,\n+ ) -> PagedList[Webhook]:\n+ \"\"\"\n+ List webhooks.\n \n+ Args:\n+ max_results: Maximum number of webhooks to return.\n+ page_token: Token specifying the next page of results.\n \n+ Returns:\n+ A :py:class:`mlflow.store.entities.paged_list.PagedList` of Webhook objects.\n+ \"\"\"\n+ return self._get_registry_client().list_webhooks(max_results, page_token)\n \n+ @experimental(version=\"3.3.0\")\n+ def update_webhook(\n+ self,\n+ webhook_id: str,\n+ name: str | None = None,\n+ description: str | None = None,\n+ url: str | None = None,\n+ events: list[WebhookEventStr | WebhookEvent] | None = None,\n+ secret: str | None = None,\n+ status: str | WebhookStatus | None = None,\n+ ) -> Webhook:\n+ \"\"\"\n+ Update an existing webhook.\n \n+ Args:\n+ webhook_id: Webhook ID.\n+ name: New webhook name.\n+ description: New webhook description.\n+ url: New webhook URL.\n+ events: New list of events. Can be strings or `WebhookEvent` objects.\n+ secret: New webhook secret.\n+ status: New webhook status. Can be string or WebhookStatus enum.\n+ Valid statuses: \"ACTIVE\", \"DISABLED\"\n \n+ Returns:\n+ A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.\n+ \"\"\"\n+ if events is not None:\n+ events = [WebhookEvent.from_str(e) if isinstance(e, str) else e for e in events]\n \n+ if status is not None:\n+ status = WebhookStatus(status) if isinstance(status, str) else status\n \n+ return self._get_registry_client().update_webhook(\n+ webhook_id=webhook_id,\n+ name=name,\n+ description=description,\n+ url=url,\n+ events=events,\n+ secret=secret,\n+ status=status,\n+ )\n \n+ @experimental(version=\"3.3.0\")\n+ def delete_webhook(self, webhook_id: str) -> None:\n+ \"\"\"\n+ Delete a webhook.\n \n+ Args:\n+ webhook_id: Webhook ID to delete.\n \n+ Returns:\n+ None\n+ \"\"\"\n+ self._get_registry_client().delete_webhook(webhook_id)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @experimental(version=\"3.3.0\")\n def test_webhook(\n@@ -6449,4 +6449,4 @@ class MlflowClient:\n Returns:\n WebhookTestResult indicating success/failure and response details.\n \"\"\"\n+ return self._get_registry_client().test_webhook(webhook_id, event)\n- return self._get_registry_client().test_webhook(webhook_id, event)\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..ed3721f8a 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,7 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..60c036051 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,36 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..b82738e23 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,12 +43,12 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+)\n-\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, extract_api_info_for_service\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n _METHOD_TO_INFO = extract_api_info_for_service(UcModelRegistryService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/utils/credentials.py b/mlflow/utils/credentials.py\nindex e8ae8d169..77f6c5cd4 100644\n+++ a/mlflow/utils/credentials.py\n--- b/mlflow/utils/credentials.py\n@@ -58,19 +58,19 @@ def read_mlflow_creds() -> MlflowCreds:\n )\n \n \n+def get_default_host_creds(store_uri):\n+ creds = read_mlflow_creds()\n+ return MlflowHostCreds(\n+ host=store_uri,\n+ username=creds.username,\n+ password=creds.password,\n+ token=MLFLOW_TRACKING_TOKEN.get(),\n+ aws_sigv4=MLFLOW_TRACKING_AWS_SIGV4.get(),\n+ auth=MLFLOW_TRACKING_AUTH.get(),\n+ ignore_tls_verification=MLFLOW_TRACKING_INSECURE_TLS.get(),\n+ client_cert_path=MLFLOW_TRACKING_CLIENT_CERT_PATH.get(),\n+ server_cert_path=MLFLOW_TRACKING_SERVER_CERT_PATH.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def login(backend: str = \"databricks\", interactive: bool = True) -> None:\n@@ -228,4 +228,4 @@ def _databricks_login(interactive):\n _validate_databricks_auth()\n except Exception as e:\n # If user entered invalid auth, we will raise an error and ask users to retry.\n+ raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n- raise MlflowException(f\"`mlflow.login()` failed with error: {e}\")\n\\ No newline at end of file\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..15aec18da 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -109,24 +109,24 @@ def _merge_json_dicts(from_dict, to_dict):\n return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -193,10 +193,10 @@ def _stringify_all_experiment_ids(x):\n _stringify_all_experiment_ids(y)\n \n \n+def parse_dict(js_dict, message):\n+ \"\"\"Parses a JSON dictionary into a message proto, ignoring unknown fields in the JSON.\"\"\"\n+ _stringify_all_experiment_ids(js_dict)\n+ ParseDict(js_dict=js_dict, message=message, ignore_unknown_fields=True)\n-\n-\n-\n-\n \n \n def set_pb_value(proto: Value, value: Any):\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\ndiff --git a/mlflow/utils/request_utils.py b/mlflow/utils/request_utils.py\nindex 4be10b5af..eaf4e57aa 100644\n+++ a/mlflow/utils/request_utils.py\n--- b/mlflow/utils/request_utils.py\n@@ -98,143 +98,143 @@ def download_chunk(*, range_start, range_end, headers, download_path, http_uri):\n f.write(response.content)\n \n \n+@lru_cache(maxsize=64)\n+def _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ # To create a new Session object for each process, we use the process id as the cache key.\n+ # This is to avoid sharing the same Session object across processes, which can lead to issues\n+ # such as https://stackoverflow.com/q/3724900.\n+ _pid,\n+ respect_retry_after_header=True,\n+):\n+ \"\"\"\n+ This function should not be called directly. Instead, use `_get_request_session` below.\n+ \"\"\"\n \n+ retry_kwargs = {\n+ \"total\": max_retries,\n+ \"connect\": max_retries,\n+ \"read\": max_retries,\n+ \"redirect\": max_retries,\n+ \"status\": max_retries,\n+ \"status_forcelist\": retry_codes,\n+ \"backoff_factor\": backoff_factor,\n+ \"backoff_jitter\": backoff_jitter,\n+ \"raise_on_status\": raise_on_status,\n+ \"respect_retry_after_header\": respect_retry_after_header,\n+ }\n+ urllib3_version = Version(urllib3.__version__)\n+ if urllib3_version >= Version(\"1.26.0\"):\n+ retry_kwargs[\"allowed_methods\"] = None\n+ else:\n+ retry_kwargs[\"method_whitelist\"] = None\n+\n+ if urllib3_version < Version(\"2.0\"):\n+ retry = JitteredRetry(**retry_kwargs)\n+ else:\n+ retry = Retry(**retry_kwargs)\n+ from mlflow.environment_variables import (\n+ MLFLOW_HTTP_POOL_CONNECTIONS,\n+ MLFLOW_HTTP_POOL_MAXSIZE,\n+ )\n+\n+ adapter = HTTPAdapter(\n+ pool_connections=MLFLOW_HTTP_POOL_CONNECTIONS.get(),\n+ pool_maxsize=MLFLOW_HTTP_POOL_MAXSIZE.get(),\n+ max_retries=retry,\n+ )\n+ session = requests.Session()\n+ session.mount(\"https://\", adapter)\n+ session.mount(\"http://\", adapter)\n+ return session\n+\n+\n+def _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+):\n+ \"\"\"Returns a `Requests.Session` object for making an HTTP request.\n \n+ Args:\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n \n+ Returns:\n+ requests.Session object.\n \n+ \"\"\"\n+ return _cached_get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ _pid=os.getpid(),\n+ respect_retry_after_header=respect_retry_after_header,\n+ )\n \n \n+def _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status=True,\n+ allow_redirects=None,\n+ respect_retry_after_header=True,\n+ **kwargs,\n+):\n+ \"\"\"Performs an HTTP request using Python's `requests` module with an automatic retry policy.\n \n+ Args:\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ url: The target URL address for the HTTP request.\n+ max_retries: Maximum total number of retries.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ session = _get_request_session(\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ respect_retry_after_header,\n+ )\n \n+ # the environment variable is hardcoded here to avoid importing mlflow.\n+ # however, documentation is available in environment_variables.py\n+ env_value = os.getenv(\"MLFLOW_ALLOW_HTTP_REDIRECTS\", \"true\").lower() in [\"true\", \"1\"]\n+ allow_redirects = env_value if allow_redirects is None else allow_redirects\n \n+ return session.request(method, url, allow_redirects=allow_redirects, **kwargs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def cloud_storage_http_request(\n@@ -276,4 +276,4 @@ def cloud_storage_http_request(\n retry_codes,\n timeout=timeout,\n **kwargs,\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..9d4bc4f06 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -32,14 +32,14 @@ from mlflow.exceptions import (\n )\n from mlflow.protos import databricks_pb2\n from mlflow.protos.databricks_pb2 import ENDPOINT_NOT_FOUND, ErrorCode\n+from mlflow.utils.proto_json_utils import parse_dict\n+from mlflow.utils.request_utils import (\n+ _TRANSIENT_FAILURE_RESPONSE_CODES,\n+ _get_http_response_with_retries,\n+ augmented_raise_for_status, # noqa: F401\n+ cloud_storage_http_request, # noqa: F401\n+)\n+from mlflow.utils.string_utils import strip_suffix\n-\n-from mlflow.utils.request_utils import _TRANSIENT_FAILURE_RESPONSE_CODES, augmented_raise_for_status, cloud_storage_http_request\n-\n-\n-\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,206 +57,206 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n \n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n \n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n \n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n+\n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n \n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n+\n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n+\n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n+\n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n+\n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n+\n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n+\n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=5)\n@@ -286,11 +286,11 @@ def get_workspace_client(\n return WorkspaceClient(config=config)\n \n \n+def _can_parse_as_json_object(string):\n+ try:\n+ return isinstance(json.loads(string), dict)\n+ except Exception:\n+ return False\n-\n-\n-\n-\n-\n \n \n def http_request_safe(host_creds, endpoint, method, **kwargs):\n@@ -301,87 +301,87 @@ def http_request_safe(host_creds, endpoint, method, **kwargs):\n return verify_rest_response(response, endpoint)\n \n \n+def verify_rest_response(response, endpoint):\n+ \"\"\"Verify the return code and format, raise exception if the request was not successful.\"\"\"\n+ # Handle Armeria-specific response case where response text is \"200 OK\"\n+ # v1/traces endpoint might return empty response\n+ if response.status_code == 200 and response.text.strip() in (_ARMERIA_OK, \"\"):\n+ response._content = b\"{}\" # Update response content to be an empty JSON dictionary\n+ return response\n+\n+ # Handle non-200 status codes\n+ if response.status_code != 200:\n+ if _can_parse_as_json_object(response.text):\n+ raise RestException(json.loads(response.text))\n+ else:\n+ base_msg = (\n+ f\"API request to endpoint {endpoint} \"\n+ f\"failed with error code {response.status_code} != 200\"\n+ )\n+ raise MlflowException(\n+ f\"{base_msg}. Response body: '{response.text}'\",\n+ error_code=get_error_code(response.status_code),\n+ )\n \n+ # Skip validation for endpoints (e.g. DBFS file-download API) which may return a non-JSON\n+ # response\n+ if endpoint.startswith(_REST_API_PATH_PREFIX) and not _can_parse_as_json_object(response.text):\n+ base_msg = (\n+ \"API request to endpoint was successful but the response body was not \"\n+ \"in a valid JSON format\"\n+ )\n+ raise MlflowException(f\"{base_msg}. Response body: '{response.text}'\")\n+\n+ return response\n+\n+\n+def _validate_max_retries(max_retries):\n+ max_retry_limit = _MLFLOW_HTTP_REQUEST_MAX_RETRIES_LIMIT.get()\n+\n+ if max_retry_limit < 0:\n+ raise MlflowException(\n+ message=f\"The current maximum retry limit is invalid ({max_retry_limit}). \"\n+ \"Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if max_retries >= max_retry_limit:\n+ raise MlflowException(\n+ message=f\"The configured max_retries value ({max_retries}) is \"\n+ f\"in excess of the maximum allowable retries ({max_retry_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if max_retries < 0:\n+ raise MlflowException(\n+ message=f\"The max_retries value must be either 0 a positive integer. Got {max_retries}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+\n+def _validate_backoff_factor(backoff_factor):\n+ max_backoff_factor_limit = _MLFLOW_HTTP_REQUEST_MAX_BACKOFF_FACTOR_LIMIT.get()\n+\n+ if max_backoff_factor_limit < 0:\n+ raise MlflowException(\n+ message=\"The current maximum backoff factor limit is invalid \"\n+ f\"({max_backoff_factor_limit}). Cannot be negative.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor >= max_backoff_factor_limit:\n+ raise MlflowException(\n+ message=f\"The configured backoff_factor value ({backoff_factor}) is in excess \"\n+ \"of the maximum allowable backoff_factor limit \"\n+ f\"({max_backoff_factor_limit})\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+\n+ if backoff_factor < 0:\n+ raise MlflowException(\n+ message=\"The backoff_factor value must be either 0 a positive integer. \"\n+ f\"Got {backoff_factor}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n@@ -565,46 +565,46 @@ def get_trace_tag_endpoint(trace_id):\n return f\"{_REST_API_PATH_PREFIX}/mlflow/traces/{trace_id}/tags\"\n \n \n+def call_endpoint(\n+ host_creds,\n+ endpoint,\n+ method,\n+ json_body,\n+ response_proto,\n+ extra_headers=None,\n+ retry_timeout_seconds=None,\n+):\n+ # Convert json string to json dictionary, to pass to requests\n+ if json_body is not None:\n+ json_body = json.loads(json_body)\n+ call_kwargs = {\n+ \"host_creds\": host_creds,\n+ \"endpoint\": endpoint,\n+ \"method\": method,\n+ }\n+ if extra_headers is not None:\n+ call_kwargs[\"extra_headers\"] = extra_headers\n+ if retry_timeout_seconds is not None:\n+ call_kwargs[\"retry_timeout_seconds\"] = retry_timeout_seconds\n+ if method == \"GET\":\n+ call_kwargs[\"params\"] = json_body\n+ response = http_request(**call_kwargs)\n+ else:\n+ call_kwargs[\"json\"] = json_body\n+ response = http_request(**call_kwargs)\n \n+ response = verify_rest_response(response, endpoint)\n+ response_to_parse = response.text\n+ try:\n+ js_dict = json.loads(response_to_parse)\n+ except json.JSONDecodeError:\n+ if len(response_to_parse) > 50:\n+ response_to_parse = response_to_parse[:50] + \"...\"\n+ _logger.warning(f\"Response is not a valid JSON object: {response_to_parse}\")\n+ raise\n \n+ parse_dict(js_dict=js_dict, message=response_proto)\n+ return response_proto\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def call_endpoints(host_creds, endpoints, json_body, response_proto, extra_headers=None):\n@@ -656,53 +656,53 @@ class MlflowHostCreds:\n client_secret: The client secret used by Databricks OAuth\n \"\"\"\n \n+ def __init__(\n+ self,\n+ host,\n+ username=None,\n+ password=None,\n+ token=None,\n+ aws_sigv4=False,\n+ auth=None,\n+ ignore_tls_verification=False,\n+ client_cert_path=None,\n+ server_cert_path=None,\n+ use_databricks_sdk=False,\n+ databricks_auth_profile=None,\n+ client_id=None,\n+ client_secret=None,\n+ use_secret_scope_token=False,\n+ ):\n+ if not host:\n+ raise MlflowException(\n+ message=\"host is a required parameter for MlflowHostCreds\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ if ignore_tls_verification and (server_cert_path is not None):\n+ raise MlflowException(\n+ message=(\n+ \"When 'ignore_tls_verification' is true then 'server_cert_path' \"\n+ \"must not be set! This error may have occurred because the \"\n+ \"'MLFLOW_TRACKING_INSECURE_TLS' and 'MLFLOW_TRACKING_SERVER_CERT_PATH' \"\n+ \"environment variables are both set - only one of these environment \"\n+ \"variables may be set.\"\n+ ),\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ self.host = host\n+ self.username = username\n+ self.password = password\n+ self.token = token\n+ self.aws_sigv4 = aws_sigv4\n+ self.auth = auth\n+ self.ignore_tls_verification = ignore_tls_verification\n+ self.client_cert_path = client_cert_path\n+ self.server_cert_path = server_cert_path\n+ self.use_databricks_sdk = use_databricks_sdk\n+ self.databricks_auth_profile = databricks_auth_profile\n+ self.client_id = client_id\n+ self.client_secret = client_secret\n+ self.use_secret_scope_token = use_secret_scope_token\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __eq__(self, other):\n if isinstance(other, self.__class__):\n@@ -712,11 +712,12 @@ class MlflowHostCreds:\n def __hash__(self):\n return hash(frozenset(self.__dict__.items()))\n \n+ @property\n+ def verify(self):\n+ if self.use_databricks_sdk:\n+ # Let databricks-sdk set HTTP request `verify` param.\n+ return None\n+ if self.server_cert_path is None:\n+ return not self.ignore_tls_verification\n+ else:\n+ return self.server_cert_path\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py\nindex e1ef46902..def9d64e8 100644\n--- a/mlflow/tracking/_model_registry/client.py\n+++ b/mlflow/tracking/_model_registry/client.py\n@@ -892,6 +892,9 @@ class ModelRegistryClient:\n Returns:\n None\n \"\"\"\n+ if not webhook_id:\n+ # No webhook ID provided, nothing to delete\n+ return\n self.store.delete_webhook(webhook_id)\n \n @experimental(version=\"3.3.0\")\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py\ndeleted file mode 100644\nindex 97fd61b00..000000000\n--- a/tests/tracking/test_client_webhooks.py\n+++ /dev/null\n@@ -1,219 +0,0 @@\n-from pathlib import Path\n-from typing import Iterator\n-\n-import pytest\n-from cryptography.fernet import Fernet\n-\n-from mlflow.entities.webhook import WebhookAction, WebhookEntity, WebhookEvent, WebhookStatus\n-from mlflow.environment_variables import MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY\n-from mlflow.exceptions import MlflowException\n-from mlflow.server import handlers\n-from mlflow.server.fastapi_app import app\n-from mlflow.server.handlers import initialize_backend_stores\n-from mlflow.tracking import MlflowClient\n-\n-from tests.helper_functions import get_safe_port\n-from tests.tracking.integration_test_utils import ServerThread\n-\n-\n-@pytest.fixture\n-def client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Iterator[MlflowClient]:\n- \"\"\"Setup a local MLflow server with proper webhook encryption key support.\"\"\"\n- # Set up encryption key for webhooks using monkeypatch\n- encryption_key = Fernet.generate_key().decode(\"utf-8\")\n- monkeypatch.setenv(MLFLOW_WEBHOOK_SECRET_ENCRYPTION_KEY.name, encryption_key)\n-\n- # Configure backend stores\n- backend_uri = f\"sqlite:///{tmp_path / 'mlflow.db'}\"\n- default_artifact_root = tmp_path.as_uri()\n-\n- # Force-reset backend stores before each test\n- handlers._tracking_store = None\n- handlers._model_registry_store = None\n- initialize_backend_stores(backend_uri, default_artifact_root=default_artifact_root)\n-\n- # Start server and return client\n- with ServerThread(app, get_safe_port()) as url:\n- yield MlflowClient(url)\n-\n-\n-def test_create_webhook(client: MlflowClient):\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n- assert webhook.events == [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n-\n- webhook = client.get_webhook(webhook.webhook_id)\n- assert webhook.name == \"test_webhook\"\n- assert webhook.url == \"https://example.com/webhook\"\n- assert webhook.secret is None\n-\n- # With secret\n- webhook_with_secret = client.create_webhook(\n- name=\"test_webhook_with_secret\",\n- url=\"https://example.com/webhook_with_secret\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- secret=\"my_secret\",\n- )\n- assert webhook_with_secret.name == \"test_webhook_with_secret\"\n- assert webhook_with_secret.url == \"https://example.com/webhook_with_secret\"\n- assert webhook_with_secret.secret is None\n- assert webhook_with_secret.events == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)\n- ]\n-\n- # Multiple events\n- webhook_multiple_events = client.create_webhook(\n- name=\"test_webhook_multiple_events\",\n- url=\"https://example.com/webhook_multiple_events\",\n- events=[\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- ],\n- )\n- assert webhook_multiple_events.name == \"test_webhook_multiple_events\"\n- assert webhook_multiple_events.url == \"https://example.com/webhook_multiple_events\"\n- assert sorted(\n- webhook_multiple_events.events, key=lambda e: (e.entity.value, e.action.value)\n- ) == [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.MODEL_VERSION_ALIAS, WebhookAction.CREATED),\n- ]\n- assert webhook_multiple_events.secret is None\n-\n-\n-def test_get_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- created_webhook = client.create_webhook(\n- name=\"test_webhook\", url=\"https://example.com/webhook\", events=events\n- )\n- retrieved_webhook = client.get_webhook(created_webhook.webhook_id)\n- assert retrieved_webhook.webhook_id == created_webhook.webhook_id\n- assert retrieved_webhook.name == \"test_webhook\"\n- assert retrieved_webhook.url == \"https://example.com/webhook\"\n- assert retrieved_webhook.events == events\n-\n-\n-def test_get_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.get_webhook(\"nonexistent\")\n-\n-\n-def test_list_webhooks(client: MlflowClient):\n- # Create more webhooks than max_results\n- created_webhooks = []\n- for i in range(5):\n- webhook = client.create_webhook(\n- name=f\"webhook{i}\",\n- url=f\"https://example.com/{i}\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- created_webhooks.append(webhook)\n- # Test pagination with max_results=2\n- webhooks_page = client.list_webhooks(max_results=2)\n- assert len(webhooks_page) == 2\n- assert webhooks_page.token is not None\n- # Get next page\n- next_webhooks_page = client.list_webhooks(max_results=2, page_token=webhooks_page.token)\n- assert len(next_webhooks_page) == 2\n- assert next_webhooks_page.token is not None\n- # Verify we don't get duplicates\n- first_page_ids = {w.webhook_id for w in webhooks_page}\n- second_page_ids = {w.webhook_id for w in next_webhooks_page}\n- assert first_page_ids.isdisjoint(second_page_ids)\n-\n-\n-def test_update_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\", url=\"https://example.com/original\", events=events\n- )\n- # Update webhook\n- new_events = [\n- WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED),\n- WebhookEvent(WebhookEntity.REGISTERED_MODEL, WebhookAction.CREATED),\n- ]\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- url=\"https://example.com/updated\",\n- events=new_events,\n- description=\"Updated description\",\n- secret=\"new_secret\",\n- status=WebhookStatus.DISABLED,\n- )\n- assert updated_webhook.webhook_id == webhook.webhook_id\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/updated\"\n- assert updated_webhook.events == new_events\n- assert updated_webhook.description == \"Updated description\"\n- assert updated_webhook.status == WebhookStatus.DISABLED\n- assert updated_webhook.last_updated_timestamp > webhook.last_updated_timestamp\n-\n-\n-def test_update_webhook_partial(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"original_name\",\n- url=\"https://example.com/original\",\n- events=events,\n- description=\"Original description\",\n- )\n- # Update only the name\n- updated_webhook = client.update_webhook(\n- webhook_id=webhook.webhook_id,\n- name=\"updated_name\",\n- )\n- assert updated_webhook.name == \"updated_name\"\n- assert updated_webhook.url == \"https://example.com/original\"\n- assert updated_webhook.events == events\n- assert updated_webhook.description == \"Original description\"\n-\n-\n-def test_update_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.update_webhook(webhook_id=\"nonexistent\", name=\"new_name\")\n-\n-\n-@pytest.mark.parametrize(\n- (\"invalid_url\", \"expected_match\"),\n- [\n- (\" \", r\"Webhook URL cannot be empty or just whitespace\"),\n- (\"ftp://example.com\", r\"Invalid webhook URL scheme\"),\n- (\"http://[invalid\", r\"Invalid webhook URL\"),\n- ],\n-)\n-def test_update_webhook_invalid_urls(client: MlflowClient, invalid_url: str, expected_match: str):\n- # Create a valid webhook first\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=[WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)],\n- )\n- with pytest.raises(MlflowException, match=expected_match):\n- client.update_webhook(webhook_id=webhook.webhook_id, url=invalid_url)\n-\n-\n-def test_delete_webhook(client: MlflowClient):\n- events = [WebhookEvent(WebhookEntity.MODEL_VERSION, WebhookAction.CREATED)]\n- webhook = client.create_webhook(\n- name=\"test_webhook\",\n- url=\"https://example.com/webhook\",\n- events=events,\n- )\n- client.delete_webhook(webhook.webhook_id)\n- with pytest.raises(MlflowException, match=r\"Webhook with ID .* not found\"):\n- client.get_webhook(webhook.webhook_id)\n- webhooks_page = client.list_webhooks()\n- webhook_ids = {w.webhook_id for w in webhooks_page}\n- assert webhook.webhook_id not in webhook_ids\n-\n-\n-def test_delete_webhook_not_found(client: MlflowClient):\n- with pytest.raises(MlflowException, match=\"Webhook with ID nonexistent not found\"):\n- client.delete_webhook(\"nonexistent\")\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S13\",\n \"shared_functions\": 2,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":63,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S22\",\n \"text\": \"MlflowException: If the request fails due to timeout, invalid URL, or other exceptions. Also raised if max_retries or backoff_factor exceed their configured limits, or if OAuth authentication is attempted without enabling Databricks SDK.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"244-253\",\n \"function\": \"http_request\"\n },\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"256-257\",\n \"function\": \"http_request\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..d376be9eb 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -248,11 +248,7 @@ def http_request(\n **kwargs,\n )\n except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n+ raise\n except requests.exceptions.InvalidURL as iu:\n raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n except Exception as e:\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S26\",\n \"shared_functions\": 1,\n \"shared_lines\": 12\n },\n {\n \"spec_id\": \"S8\",\n \"shared_functions\": 1,\n \"shared_lines\": 12\n },\n {\n \"spec_id\": \"S30\",\n \"shared_functions\": 1,\n \"shared_lines\": 5\n }\n ]\n}"}}},{"rowIdx":64,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S23\",\n \"text\": \"InvalidUrlException: If the constructed URL is invalid.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"254-255\",\n \"function\": \"http_request\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..0109d1c44 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -254,7 +254,7 @@ def http_request(\n f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n ) from to\n except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ raise MlflowException(f\"Invalid url: {url}\") from iu\n except Exception as e:\n raise MlflowException(f\"API request to {url} failed with exception {e}\")\n \n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S26\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S8\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":65,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S26\",\n \"text\": \"Authentication is handled automatically based on the credentials provided in host_creds (username/password, token, client certificates, AWS SigV4, etc.).\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"239-268\",\n \"function\": \"http_request\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..516930bee 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -207,8 +207,6 @@ def http_request(\n if host_creds.username and host_creds.password:\n basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n elif host_creds.client_secret:\n raise MlflowException(\n \"To use OAuth authentication, set environmental variable \"\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S8\",\n \"shared_functions\": 1,\n \"shared_lines\": 24\n },\n {\n \"spec_id\": \"S22\",\n \"shared_functions\": 1,\n \"shared_lines\": 12\n },\n {\n \"spec_id\": \"S30\",\n \"shared_functions\": 1,\n \"shared_lines\": 5\n },\n {\n \"spec_id\": \"S23\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S48\",\n \"shared_functions\": 0,\n \"shared_lines\": 4\n }\n ]\n}"}}},{"rowIdx":66,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S31\",\n \"text\": \"Validate that total retry timeout is not less than single request timeout. This function performs a validation check to ensure that the total retry timeout duration is not configured to be shorter than the timeout for a single HTTP request. When the total retry timeout is less than the single request timeout, it can cause premature failures where the retry mechanism times out before a single long-running request has a chance to complete.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"387-407\",\n \"function\": \"validate_deployment_timeout_config\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..d270b8bec 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -393,7 +393,7 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second\n retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n \"\"\"\n if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n+ if retry_timeout_seconds <= timeout:\n warnings.warn(\n f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S36\",\n \"shared_functions\": 1,\n \"shared_lines\": 9\n },\n {\n \"spec_id\": \"S38\",\n \"shared_functions\": 1,\n \"shared_lines\": 8\n },\n {\n \"spec_id\": \"S35\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S32\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S33\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S34\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S37\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n }\n ]\n}"}}},{"rowIdx":67,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S32\",\n \"text\": \"timeout (int | None): Maximum time in seconds to wait for a single HTTP request to complete. If None, no single request timeout validation is performed.\",\n \"category\": \"structural\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"387\",\n \"function\": \"validate_deployment_timeout_config\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..e066e8dc8 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -392,7 +392,14 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second\n timeout: Maximum time for a single HTTP request (in seconds)\n retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n+ if timeout is None and retry_timeout_seconds is not None:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ \"but MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT is not set. Consider setting \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT for better control over request timeouts.\",\n+ stacklevel=2,\n+ )\n+ elif timeout is not None and retry_timeout_seconds is not None:\n if retry_timeout_seconds < timeout:\n warnings.warn(\n f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S33\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S34\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S35\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":68,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S33\",\n \"text\": \"retry_timeout_seconds (int | None): Maximum time in seconds to spend on all retry attempts combined. If None, no retry timeout validation is performed.\",\n \"category\": \"structural\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"387\",\n \"function\": \"validate_deployment_timeout_config\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..d8ae255b0 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -392,8 +392,8 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second\n timeout: Maximum time for a single HTTP request (in seconds)\n retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n+ if timeout is not None:\n+ if retry_timeout_seconds is None or retry_timeout_seconds < timeout:\n warnings.warn(\n f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S32\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S34\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":69,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S34\",\n \"text\": \"None: This function does not return any value. It only performs validation and may issue warnings.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"407\",\n \"function\": \"validate_deployment_timeout_config\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..8e70ecb9e 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -403,6 +403,7 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second\n f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n stacklevel=2,\n )\n+ return True\n \n \n def _time_sleep(seconds: float) -> None:\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S32\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S33\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":70,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S37\",\n \"text\": \"The function uses stacklevel=2 in the warning to point to the caller's location rather than this validation function\",\n \"category\": \"implementation_constraint\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"405\",\n \"function\": \"validate_deployment_timeout_config\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..1cf0dbb1b 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -401,7 +401,7 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second\n \"premature failures. For long-running predictions, ensure \"\n \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n+ stacklevel=1,\n )\n \n \n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S36\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S38\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n }\n ]\n}"}}},{"rowIdx":71,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S38\",\n \"text\": \"The warning suggests setting MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least the value of MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"399-406\",\n \"function\": \"validate_deployment_timeout_config\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..b630377d1 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -400,7 +400,7 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second\n \"total retry timeout could expire before a single request completes, causing \"\n \"premature failures. For long-running predictions, ensure \"\n \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout - 10}s.\",\n stacklevel=2,\n )\n \n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 1,\n \"shared_lines\": 8\n },\n {\n \"spec_id\": \"S36\",\n \"shared_functions\": 1,\n \"shared_lines\": 8\n },\n {\n \"spec_id\": \"S37\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n }\n ]\n}"}}},{"rowIdx":72,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S48\",\n \"text\": \"get_workspace_client creates and caches WorkspaceClient instances with specified configuration including retry and timeout settings\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/rest_utils.py\",\n \"lines\": \"265-289\",\n \"function\": \"get_workspace_client\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n--- a/mlflow/deployments/__init__.py\n+++ b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n-from mlflow.deployments.interface import get_deploy_client, run_local\n+from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n--- a/mlflow/deployments/base.py\n+++ b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n- def __init__(self, target_uri):\n- self.target_uri = target_uri\n+\n+\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n--- a/mlflow/deployments/databricks/__init__.py\n+++ b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.rest_utils import (\n- augmented_raise_for_status,\n- http_request,\n- validate_deployment_timeout_config,\n-)\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n+\n+\n+\n+\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n--- a/mlflow/deployments/interface.py\n+++ b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n-from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n+from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n-def get_deploy_client(target_uri=None):\n- \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n- APIs for deploying models to the specified target. See available deployment APIs\n- by calling ``help()`` on the returned object or viewing docs for\n- :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n- ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n- configuration options.\n \n- Args:\n- target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n- MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n- `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n-\n- .. code-block:: python\n- :caption: Example\n-\n- from mlflow.deployments import get_deploy_client\n- import pandas as pd\n-\n- client = get_deploy_client(\"redisai\")\n- # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n- # model artifacts are fetched from the current tracking server and then used for deployment.\n- client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n- # Load a CSV of emails and score it against our deployment\n- emails_df = pd.read_csv(\"...\")\n- prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n- # List all deployments, get details of our particular deployment\n- print(client.list_deployments())\n- print(client.get_deployment(\"spamDetector\"))\n- # Update our deployment to serve a different model\n- client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n- # Delete our deployment\n- client.delete_deployment(\"spamDetector\")\n- \"\"\"\n- if not target_uri:\n- try:\n- target_uri = get_deployments_target()\n- except MlflowException:\n- _logger.info(\n- \"No deployments target has been set. Please either set the MLflow deployments \"\n- \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n- \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n- )\n- return None\n- target = parse_target_uri(target_uri)\n- plugin = plugin_store[target]\n- for _, obj in inspect.getmembers(plugin):\n- if inspect.isclass(obj):\n- if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n- return obj(target_uri)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n- return plugin_store[target].target_help()\n+ return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n--- a/mlflow/deployments/mlflow/__init__.py\n+++ b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n--- a/mlflow/deployments/plugin_manager.py\n+++ b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n-from mlflow.deployments.utils import parse_target_uri\n+\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n- @property\n- def registry(self):\n- \"\"\"\n- Registry stores the registered plugin as a key value pair where key is the\n- name of the plugin and value is the plugin object\n- \"\"\"\n- return self._registry\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n- return plugin_obj\n+ return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n--- a/mlflow/deployments/utils.py\n+++ b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n-def parse_target_uri(target_uri):\n- \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n- parsed = urllib.parse.urlparse(target_uri)\n- if not parsed.scheme:\n- if parsed.path:\n- # uri = 'target_name' (without :/<path>)\n- return parsed.path\n- raise MlflowException(\n- f\"Not a proper deployment URI: {target_uri}. \"\n- + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n- )\n- return parsed.scheme\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n- return _is_valid_uri(target)\n+ return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- call_endpoint,\n- extract_all_api_info_for_service,\n- extract_api_info_for_service,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n--- a/mlflow/store/artifact/databricks_models_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n-from mlflow.utils.rest_utils import http_request\n+\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n--- a/mlflow/store/artifact/dbfs_artifact_repo.py\n+++ b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n-from mlflow.utils.rest_utils import (\n- RESOURCE_NON_EXISTENT,\n- http_request,\n- http_request_safe,\n-)\n+from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n+\n+\n+\n+\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n--- a/mlflow/store/artifact/http_artifact_repo.py\n+++ b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n-from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n+from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V4_REST_API_PATH_PREFIX,\n- _V4_TRACE_REST_API_PATH_PREFIX,\n- extract_api_info_for_service,\n- get_single_assessment_endpoint_v4,\n- get_single_trace_endpoint_v4,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n-from mlflow.utils.rest_utils import (\n- _REST_API_PATH_PREFIX,\n- _V3_TRACE_REST_API_PATH_PREFIX,\n- MlflowHostCreds,\n- call_endpoint,\n- extract_api_info_for_service,\n- get_logged_model_endpoint,\n- get_single_assessment_endpoint,\n- get_single_trace_endpoint,\n- get_trace_tag_endpoint,\n- http_request,\n- verify_rest_response,\n-)\n+from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n--- a/mlflow/tracking/request_header/databricks_request_header_provider.py\n+++ b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n- def in_context(self):\n- return (\n- databricks_utils.is_in_cluster()\n- or databricks_utils.is_in_databricks_notebook()\n- or databricks_utils.is_in_databricks_job()\n- )\n+\n+\n+\n+\n+\n+\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n- return request_headers\n+ return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n--- a/mlflow/tracking/request_header/default_request_header_provider.py\n+++ b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,8 +15,6 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n- def in_context(self):\n- return True\n \n- def request_headers(self):\n- return dict(**_DEFAULT_HEADERS)\n+\n+\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n--- a/mlflow/tracking/request_header/registry.py\n+++ b/mlflow/tracking/request_header/registry.py\n@@ -43,37 +43,35 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n-def resolve_request_headers(request_headers=None):\n- \"\"\"Generate a set of request headers from registered providers.\n-\n- Request headers are resolved in the order that providers are registered. Argument headers are\n- applied last. This function iterates through all request header providers in the registry.\n- Additional context providers can be registered as described in\n- :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n-\n- Args:\n- request_headers: A dictionary of request headers to override. If specified, headers passed\n- in this argument will override those inferred from the context.\n-\n- Returns:\n- A dictionary of resolved headers.\n- \"\"\"\n-\n- all_request_headers = {}\n- for provider in _request_header_provider_registry:\n- try:\n- if provider.in_context():\n- # all_request_headers.update(provider.request_headers())\n- for header, value in provider.request_headers().items():\n- all_request_headers[header] = (\n- f\"{all_request_headers[header]} {value}\"\n- if header in all_request_headers\n- else value\n- )\n- except Exception as e:\n- _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n-\n- if request_headers is not None:\n- all_request_headers.update(request_headers)\n-\n- return all_request_headers\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n--- a/mlflow/utils/databricks_utils.py\n+++ b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n-from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n+from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n-from mlflow.utils.string_utils import strip_suffix\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n-def http_request(\n- host_creds,\n- endpoint,\n- method,\n- max_retries=None,\n- backoff_factor=None,\n- backoff_jitter=None,\n- extra_headers=None,\n- retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n- timeout=None,\n- raise_on_status=True,\n- respect_retry_after_header=None,\n- retry_timeout_seconds=None,\n- **kwargs,\n-):\n- \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n- errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n- retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n- The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n- Args:\n- host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n- hostname and optional authentication.\n- endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n- method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n- max_retries: Maximum number of retries before throwing an exception.\n- backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n- request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n- exponential backoff.\n- backoff_jitter: A random jitter to add to the backoff interval.\n- extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n- retry_codes: A list of HTTP response error codes that qualifies for retry.\n- timeout: Wait for timeout seconds for response from remote server for connect and\n- read request.\n- raise_on_status: Whether to raise an exception, or return a response, if status falls\n- in retry_codes range and retries have been exhausted.\n- respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n- as Retry.RETRY_AFTER_STATUS_CODES or not.\n- retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n- kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n-\n- Returns:\n- requests.Response object.\n- \"\"\"\n- cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n- url = f\"{cleaned_hostname}{endpoint}\"\n-\n- # Set defaults for retry parameters from environment variables if not specified\n- max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n- backoff_factor = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n- )\n- backoff_jitter = (\n- MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n- )\n-\n- from mlflow.tracking.request_header.registry import resolve_request_headers\n-\n- headers = dict(**resolve_request_headers())\n- if extra_headers:\n- headers = dict(**headers, **extra_headers)\n-\n- if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n- headers[\"x-databricks-traffic-id\"] = traffic_id\n-\n- if host_creds.use_databricks_sdk:\n- from databricks.sdk.errors import DatabricksError\n-\n- ws_client = get_workspace_client(\n- host_creds.use_secret_scope_token,\n- host_creds.host,\n- host_creds.token,\n- host_creds.databricks_auth_profile,\n- retry_timeout_seconds=retry_timeout_seconds,\n- timeout=timeout,\n- )\n \n- def make_sdk_call():\n- # Databricks SDK `APIClient.do` API is for making request using\n- # HTTP\n- # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n- # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n- with warnings.catch_warnings():\n- warnings.filterwarnings(\n- \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n- )\n- raw_response = ws_client.api_client.do(\n- method=method,\n- path=endpoint,\n- headers=headers,\n- raw=True,\n- query=kwargs.get(\"params\"),\n- body=kwargs.get(\"json\"),\n- files=kwargs.get(\"files\"),\n- data=kwargs.get(\"data\"),\n- )\n- return raw_response[\"contents\"]._response\n \n- try:\n- # We retry the SDK call with exponential backoff because the Databricks SDK default\n- # retry behavior does not handle all transient errors that we want to retry, and it\n- # does not support a customizable retry policy based on HTTP response status codes.\n- # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n- # response strings that Databricks SDK retries on), the SDK may retry internally,\n- # and MLflow may retry on top of that, leading to more retries than specified by\n- # `max_retries`. This is acceptable, given the enforcement of an overall request\n- # timeout via `retry_timeout_seconds`.\n- #\n- # TODO: Update transient error handling defaults in Databricks SDK to match standard\n- # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n- # SDK, and remove this custom retry wrapper from MLflow\n- return _retry_databricks_sdk_call_with_exponential_backoff(\n- call_func=make_sdk_call,\n- retry_codes=retry_codes,\n- retry_timeout_seconds=(\n- retry_timeout_seconds\n- if retry_timeout_seconds is not None\n- else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n- ),\n- backoff_factor=backoff_factor,\n- backoff_jitter=backoff_jitter,\n- max_retries=max_retries,\n- )\n- except DatabricksError as e:\n- response = requests.Response()\n- response.url = url\n- response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n- response.reason = str(e)\n- response.encoding = \"UTF-8\"\n- response._content = json.dumps(\n- {\n- \"error_code\": e.error_code,\n- \"message\": str(e),\n- }\n- ).encode(\"UTF-8\")\n- return response\n-\n- _validate_max_retries(max_retries)\n- _validate_backoff_factor(backoff_factor)\n- respect_retry_after_header = (\n- MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n- if respect_retry_after_header is None\n- else respect_retry_after_header\n- )\n-\n- timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n- auth_str = None\n- if host_creds.username and host_creds.password:\n- basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n- auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n- elif host_creds.token:\n- auth_str = f\"Bearer {host_creds.token}\"\n- elif host_creds.client_secret:\n- raise MlflowException(\n- \"To use OAuth authentication, set environmental variable \"\n- f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n- error_code=CUSTOMER_UNAUTHORIZED,\n- )\n \n- if auth_str:\n- headers[\"Authorization\"] = auth_str\n \n- if host_creds.client_cert_path is not None:\n- kwargs[\"cert\"] = host_creds.client_cert_path\n \n- if host_creds.aws_sigv4:\n- # will overwrite the Authorization header\n- from requests_auth_aws_sigv4 import AWSSigV4\n \n- kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n- elif host_creds.auth:\n- from mlflow.tracking.request_auth.registry import fetch_auth\n \n- kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n- try:\n- return _get_http_response_with_retries(\n- method,\n- url,\n- max_retries,\n- backoff_factor,\n- backoff_jitter,\n- retry_codes,\n- raise_on_status,\n- headers=headers,\n- verify=host_creds.verify,\n- timeout=timeout,\n- respect_retry_after_header=respect_retry_after_header,\n- **kwargs,\n- )\n- except requests.exceptions.Timeout as to:\n- raise MlflowException(\n- f\"API request to {url} failed with timeout exception {to}.\"\n- \" To increase the timeout, set the environment variable \"\n- f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n- ) from to\n- except requests.exceptions.InvalidURL as iu:\n- raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n- except Exception as e:\n- raise MlflowException(f\"API request to {url} failed with exception {e}\")\n-\n-\n-@lru_cache(maxsize=5)\n-def get_workspace_client(\n- use_secret_scope_token,\n- host,\n- token,\n- databricks_auth_profile,\n- retry_timeout_seconds=None,\n- timeout=None,\n-):\n- from databricks.sdk import WorkspaceClient\n- from databricks.sdk.config import Config\n \n- if use_secret_scope_token:\n- kwargs = {\"host\": host, \"token\": token}\n- else:\n- kwargs = {\"profile\": databricks_auth_profile}\n- if timeout is not None:\n- kwargs[\"http_timeout_seconds\"] = timeout\n- config = Config(\n- **kwargs,\n- retry_timeout_seconds=retry_timeout_seconds\n- or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n- )\n- # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n- return WorkspaceClient(config=config)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n-def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n- \"\"\"\n- Validate that total retry timeout is not less than single request timeout.\n \n- Args:\n- timeout: Maximum time for a single HTTP request (in seconds)\n- retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n- \"\"\"\n- if timeout is not None and retry_timeout_seconds is not None:\n- if retry_timeout_seconds < timeout:\n- warnings.warn(\n- f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n- f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n- \"total retry timeout could expire before a single request completes, causing \"\n- \"premature failures. For long-running predictions, ensure \"\n- \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n- f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n- stacklevel=2,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n-def _retry_databricks_sdk_call_with_exponential_backoff(\n- *,\n- call_func: Callable[..., Any],\n- retry_codes: list[int],\n- retry_timeout_seconds: int,\n- backoff_factor: int,\n- backoff_jitter: float,\n- max_retries: int,\n-):\n- \"\"\"\n- Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n- Args:\n- call_func: Function to call that may raise DatabricksError\n- retry_codes: Set of HTTP status codes that should trigger retries\n- retry_timeout_seconds: Maximum time to spend retrying in seconds\n- backoff_factor: Factor for exponential backoff\n- backoff_jitter: Random jitter to add to backoff\n- max_retries: Maximum number of retry attempts\n-\n- Returns:\n- The result of call_func() on success\n-\n- Raises:\n- DatabricksError: If all retries are exhausted or non-retryable error occurs\n- \"\"\"\n- from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n- start_time = time.time()\n- attempt = 0\n \n- while attempt <= max_retries:\n- try:\n- return call_func()\n- except DatabricksError as e:\n- # Get HTTP status code from the error\n- status_code = next(\n- (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n- )\n- # Check if this is a retryable error\n- if status_code not in retry_codes:\n- raise\n-\n- # Check if we've exceeded max retries\n- if attempt >= max_retries:\n- _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n- raise\n-\n- # Calculate backoff time with exponential backoff and jitter\n- # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n- # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n- # maintains compatibility with, doesn't support retries with jitter\n- if attempt <= 0:\n- backoff_time = 0 # No backoff on first retry attempt\n- else:\n- backoff_time = backoff_factor * (2**attempt)\n- if backoff_jitter > 0:\n- backoff_time += random.random() * backoff_jitter\n-\n- # Check if we've exceeded or would exceed timeout\n- elapsed_time = time.time() - start_time\n- if elapsed_time + backoff_time >= retry_timeout_seconds:\n- _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n- raise\n-\n- _logger.debug(\n- f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n- f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n- )\n \n- _time_sleep(backoff_time)\n- attempt += 1\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n- return self.server_cert_path\n+ return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n--- a/mlflow/utils/string_utils.py\n+++ b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n-def strip_suffix(original: str, suffix: str) -> str:\n- if original.endswith(suffix) and suffix != \"\":\n- return original[: -len(suffix)]\n- return original\n+\n+\n+\n+\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n- return str(cell_value)\n+ return str(cell_value)\n\\ No newline at end of file\ndiff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\nindex db1f15091..ef63138c8 100644\n--- a/tests/deployments/databricks/test_databricks.py\n+++ b/tests/deployments/databricks/test_databricks.py\n@@ -4,7 +4,7 @@ from unittest import mock\n \n import pytest\n \n-from mlflow.deployments import get_deploy_client\n+\n from mlflow.exceptions import MlflowException\n \n \n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py\nindex 9d3545850..33d590fdc 100644\n+++ a/mlflow/deployments/__init__.py\n--- b/mlflow/deployments/__init__.py\n@@ -19,7 +19,7 @@ import json\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.databricks import DatabricksDeploymentClient, DatabricksEndpoint\n+from mlflow.deployments.interface import get_deploy_client, run_local\n-from mlflow.deployments.interface import run_local\n from mlflow.deployments.openai import OpenAIDeploymentClient\n from mlflow.deployments.utils import get_deployments_target, set_deployments_target\n from mlflow.exceptions import MlflowException\ndiff --git a/mlflow/deployments/base.py b/mlflow/deployments/base.py\nindex cd79f1d31..c78d4bb0b 100644\n+++ a/mlflow/deployments/base.py\n--- b/mlflow/deployments/base.py\n@@ -86,8 +86,8 @@ class BaseDeploymentClient(abc.ABC):\n on failure to deploy a model).\n \"\"\"\n \n+ def __init__(self, target_uri):\n+ self.target_uri = target_uri\n-\n-\n \n @abc.abstractmethod\n def create_deployment(self, name, model_uri, flavor=None, config=None, endpoint=None):\n@@ -355,4 +355,4 @@ class BaseDeploymentClient(abc.ABC):\n raise MlflowException(\n \"Method is unimplemented in base client. Implementation should be \"\n \"provided by specific target plugins.\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/databricks/__init__.py b/mlflow/deployments/databricks/__init__.py\nindex 7630a32b5..df4a4ea37 100644\n+++ a/mlflow/deployments/databricks/__init__.py\n--- b/mlflow/deployments/databricks/__init__.py\n@@ -16,11 +16,11 @@ from mlflow.exceptions import MlflowException\n from mlflow.utils import AttrDict\n from mlflow.utils.annotations import deprecated\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.rest_utils import (\n+ augmented_raise_for_status,\n+ http_request,\n+ validate_deployment_timeout_config,\n+)\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n-\n-\n-\n-\n \n \n class DatabricksEndpoint(AttrDict):\ndiff --git a/mlflow/deployments/interface.py b/mlflow/deployments/interface.py\nindex 3e4e6f68c..b7242ca66 100644\n+++ a/mlflow/deployments/interface.py\n--- b/mlflow/deployments/interface.py\n@@ -3,7 +3,7 @@ from logging import Logger\n \n from mlflow.deployments.base import BaseDeploymentClient\n from mlflow.deployments.plugin_manager import DeploymentPlugins\n+from mlflow.deployments.utils import get_deployments_target, parse_target_uri\n-from mlflow.deployments.utils import get_deployments_target\n from mlflow.exceptions import MlflowException\n \n plugin_store = DeploymentPlugins()\n@@ -12,56 +12,56 @@ plugin_store.register(\"sagemaker\", \"mlflow.sagemaker\")\n _logger = Logger(__name__)\n \n \n+def get_deploy_client(target_uri=None):\n+ \"\"\"Returns a subclass of :py:class:`mlflow.deployments.BaseDeploymentClient` exposing standard\n+ APIs for deploying models to the specified target. See available deployment APIs\n+ by calling ``help()`` on the returned object or viewing docs for\n+ :py:class:`mlflow.deployments.BaseDeploymentClient`. You can also run\n+ ``mlflow deployments help -t <target-uri>`` via the CLI for more details on target-specific\n+ configuration options.\n \n+ Args:\n+ target_uri: Optional URI of target to deploy to. If no target URI is provided, then\n+ MLflow will attempt to get the deployments target set via `get_deployments_target()` or\n+ `MLFLOW_DEPLOYMENTS_TARGET` environment variable.\n+\n+ .. code-block:: python\n+ :caption: Example\n+\n+ from mlflow.deployments import get_deploy_client\n+ import pandas as pd\n+\n+ client = get_deploy_client(\"redisai\")\n+ # Deploy the model stored at artifact path 'myModel' under run with ID 'someRunId'. The\n+ # model artifacts are fetched from the current tracking server and then used for deployment.\n+ client.create_deployment(\"spamDetector\", \"runs:/someRunId/myModel\")\n+ # Load a CSV of emails and score it against our deployment\n+ emails_df = pd.read_csv(\"...\")\n+ prediction_df = client.predict_deployment(\"spamDetector\", emails_df)\n+ # List all deployments, get details of our particular deployment\n+ print(client.list_deployments())\n+ print(client.get_deployment(\"spamDetector\"))\n+ # Update our deployment to serve a different model\n+ client.update_deployment(\"spamDetector\", \"runs:/anotherRunId/myModel\")\n+ # Delete our deployment\n+ client.delete_deployment(\"spamDetector\")\n+ \"\"\"\n+ if not target_uri:\n+ try:\n+ target_uri = get_deployments_target()\n+ except MlflowException:\n+ _logger.info(\n+ \"No deployments target has been set. Please either set the MLflow deployments \"\n+ \"target via `mlflow.deployments.set_deployments_target()` or set the environment \"\n+ \"variable MLFLOW_DEPLOYMENTS_TARGET to the running deployment server's uri\"\n+ )\n+ return None\n+ target = parse_target_uri(target_uri)\n+ plugin = plugin_store[target]\n+ for _, obj in inspect.getmembers(plugin):\n+ if inspect.isclass(obj):\n+ if issubclass(obj, BaseDeploymentClient) and not obj == BaseDeploymentClient:\n+ return obj(target_uri)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def run_local(target, name, model_uri, flavor=None, config=None):\n@@ -99,4 +99,4 @@ def _target_help(target):\n Args:\n target: Which target to use. This information is used to call the appropriate plugin.\n \"\"\"\n+ return plugin_store[target].target_help()\n- return plugin_store[target].target_help()\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/mlflow/__init__.py b/mlflow/deployments/mlflow/__init__.py\nindex 4ce936434..80b208c31 100644\n+++ a/mlflow/deployments/mlflow/__init__.py\n--- b/mlflow/deployments/mlflow/__init__.py\n@@ -20,7 +20,7 @@ from mlflow.environment_variables import (\n from mlflow.protos.databricks_pb2 import BAD_REQUEST\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.utils.credentials import get_default_host_creds\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import join_paths\n \n if TYPE_CHECKING:\ndiff --git a/mlflow/deployments/plugin_manager.py b/mlflow/deployments/plugin_manager.py\nindex 0cba055a0..f73392203 100644\n+++ a/mlflow/deployments/plugin_manager.py\n--- b/mlflow/deployments/plugin_manager.py\n@@ -5,7 +5,7 @@ import inspect\n import importlib_metadata\n \n from mlflow.deployments.base import BaseDeploymentClient\n+from mlflow.deployments.utils import parse_target_uri\n-\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, RESOURCE_DOES_NOT_EXIST\n from mlflow.utils.annotations import developer_stable\n@@ -38,13 +38,13 @@ class PluginManager(abc.ABC):\n # can raise custom exceptions if it needs to\n pass\n \n+ @property\n+ def registry(self):\n+ \"\"\"\n+ Registry stores the registered plugin as a key value pair where key is the\n+ name of the plugin and value is the plugin object\n+ \"\"\"\n+ return self._registry\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def has_registered(self):\n@@ -140,4 +140,4 @@ class DeploymentPlugins(PluginManager):\n \" of BaseDeploymentClient. Raise an issue with the \"\n \"plugin developers\"\n )\n+ return plugin_obj\n- return plugin_obj\n\\ No newline at end of file\ndiff --git a/mlflow/deployments/utils.py b/mlflow/deployments/utils.py\nindex a76987798..4a8dac68b 100644\n+++ a/mlflow/deployments/utils.py\n--- b/mlflow/deployments/utils.py\n@@ -8,18 +8,18 @@ from mlflow.utils.uri import append_to_uri_path\n _deployments_target: str | None = None\n \n \n+def parse_target_uri(target_uri):\n+ \"\"\"Parse out the deployment target from the provided target uri\"\"\"\n+ parsed = urllib.parse.urlparse(target_uri)\n+ if not parsed.scheme:\n+ if parsed.path:\n+ # uri = 'target_name' (without :/<path>)\n+ return parsed.path\n+ raise MlflowException(\n+ f\"Not a proper deployment URI: {target_uri}. \"\n+ + \"Deployment URIs must be of the form 'target' or 'target:/suffix'\"\n+ )\n+ return parsed.scheme\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _is_valid_uri(uri: str) -> bool:\n@@ -93,4 +93,4 @@ def _is_valid_target(target: str):\n \"\"\"\n if target == \"databricks\":\n return True\n+ return _is_valid_uri(target)\n- return _is_valid_uri(target)\n\\ No newline at end of file\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..0f388960d 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -158,14 +158,14 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ call_endpoint,\n+ extract_all_api_info_for_service,\n+ extract_api_info_for_service,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, call_endpoint, extract_all_api_info_for_service, extract_api_info_for_service, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.uri import is_fuse_or_uc_volumes_uri\n \n _TRACKING_METHOD_TO_INFO = extract_api_info_for_service(MlflowService, _REST_API_PATH_PREFIX)\ndiff --git a/mlflow/store/artifact/databricks_models_artifact_repo.py b/mlflow/store/artifact/databricks_models_artifact_repo.py\nindex 678dde650..9a995804b 100644\n+++ a/mlflow/store/artifact/databricks_models_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_models_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.utils.file_utils import (\n parallelized_download_file_using_http_uri,\n remove_on_error,\n )\n+from mlflow.utils.rest_utils import http_request\n-\n from mlflow.utils.uri import get_databricks_profile_uri_from_artifact_uri\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/artifact/dbfs_artifact_repo.py b/mlflow/store/artifact/dbfs_artifact_repo.py\nindex 4aee71b05..24b605b92 100644\n+++ a/mlflow/store/artifact/dbfs_artifact_repo.py\n--- b/mlflow/store/artifact/dbfs_artifact_repo.py\n@@ -21,11 +21,11 @@ from mlflow.store.tracking.rest_store import RestStore\n from mlflow.tracking._tracking_service import utils\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n+from mlflow.utils.rest_utils import (\n+ RESOURCE_NON_EXISTENT,\n+ http_request,\n+ http_request_safe,\n+)\n-from mlflow.utils.rest_utils import RESOURCE_NON_EXISTENT, http_request_safe\n-\n-\n-\n-\n from mlflow.utils.string_utils import strip_prefix\n from mlflow.utils.uri import (\n get_databricks_profile_uri_from_artifact_uri,\ndiff --git a/mlflow/store/artifact/http_artifact_repo.py b/mlflow/store/artifact/http_artifact_repo.py\nindex 05fa91367..d17ff164e 100644\n+++ a/mlflow/store/artifact/http_artifact_repo.py\n--- b/mlflow/store/artifact/http_artifact_repo.py\n@@ -26,7 +26,7 @@ from mlflow.store.artifact.cloud_artifact_repo import _complete_futures, _comput\n from mlflow.utils.credentials import get_default_host_creds\n from mlflow.utils.file_utils import read_chunk, relative_path_to_artifact_path\n from mlflow.utils.mime_type_utils import _guess_mime_type\n+from mlflow.utils.rest_utils import augmented_raise_for_status, http_request\n-from mlflow.utils.rest_utils import augmented_raise_for_status\n from mlflow.utils.uri import validate_path_is_safe\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..9b87af835 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -53,16 +53,16 @@ from mlflow.utils.databricks_tracing_utils import (\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V4_REST_API_PATH_PREFIX,\n+ _V4_TRACE_REST_API_PATH_PREFIX,\n+ extract_api_info_for_service,\n+ get_single_assessment_endpoint_v4,\n+ get_single_trace_endpoint_v4,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V4_REST_API_PATH_PREFIX, _V4_TRACE_REST_API_PATH_PREFIX, extract_api_info_for_service, get_single_assessment_endpoint_v4, get_single_trace_endpoint_v4, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n DATABRICKS_UC_TABLE_HEADER = \"X-Databricks-UC-Table-Name\"\n \ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..3b760bcd3 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -121,19 +121,19 @@ from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n from mlflow.utils.proto_json_utils import message_to_json\n+from mlflow.utils.rest_utils import (\n+ _REST_API_PATH_PREFIX,\n+ _V3_TRACE_REST_API_PATH_PREFIX,\n+ MlflowHostCreds,\n+ call_endpoint,\n+ extract_api_info_for_service,\n+ get_logged_model_endpoint,\n+ get_single_assessment_endpoint,\n+ get_single_trace_endpoint,\n+ get_trace_tag_endpoint,\n+ http_request,\n+ verify_rest_response,\n+)\n-from mlflow.utils.rest_utils import _REST_API_PATH_PREFIX, _V3_TRACE_REST_API_PATH_PREFIX, MlflowHostCreds, call_endpoint, extract_api_info_for_service, get_logged_model_endpoint, get_single_assessment_endpoint, get_single_trace_endpoint, get_trace_tag_endpoint, verify_rest_response\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.utils.validation import _resolve_experiment_ids_and_locations\n \n _logger = logging.getLogger(__name__)\ndiff --git a/mlflow/tracking/request_header/databricks_request_header_provider.py b/mlflow/tracking/request_header/databricks_request_header_provider.py\nindex 420e4aeca..c51ecd4b9 100644\n+++ a/mlflow/tracking/request_header/databricks_request_header_provider.py\n--- b/mlflow/tracking/request_header/databricks_request_header_provider.py\n@@ -8,12 +8,12 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n was made.\n \"\"\"\n \n+ def in_context(self):\n+ return (\n+ databricks_utils.is_in_cluster()\n+ or databricks_utils.is_in_databricks_notebook()\n+ or databricks_utils.is_in_databricks_job()\n+ )\n-\n-\n-\n-\n-\n-\n \n def request_headers(self):\n request_headers = {}\n@@ -35,4 +35,4 @@ class DatabricksRequestHeaderProvider(RequestHeaderProvider):\n if workload_class is not None:\n request_headers[\"workload_class\"] = workload_class\n \n+ return request_headers\n- return request_headers\n\\ No newline at end of file\ndiff --git a/mlflow/tracking/request_header/default_request_header_provider.py b/mlflow/tracking/request_header/default_request_header_provider.py\nindex 460bfa66f..c0e9486bb 100644\n+++ a/mlflow/tracking/request_header/default_request_header_provider.py\n--- b/mlflow/tracking/request_header/default_request_header_provider.py\n@@ -15,6 +15,8 @@ class DefaultRequestHeaderProvider(RequestHeaderProvider):\n Provides default request headers for outgoing request.\n \"\"\"\n \n+ def in_context(self):\n+ return True\n \n+ def request_headers(self):\n+ return dict(**_DEFAULT_HEADERS)\n-\n-\ndiff --git a/mlflow/tracking/request_header/registry.py b/mlflow/tracking/request_header/registry.py\nindex 9cdb2bf21..190286f5e 100644\n+++ a/mlflow/tracking/request_header/registry.py\n--- b/mlflow/tracking/request_header/registry.py\n@@ -43,35 +43,37 @@ _request_header_provider_registry.register(DefaultRequestHeaderProvider)\n _request_header_provider_registry.register_entrypoints()\n \n \n+def resolve_request_headers(request_headers=None):\n+ \"\"\"Generate a set of request headers from registered providers.\n+\n+ Request headers are resolved in the order that providers are registered. Argument headers are\n+ applied last. This function iterates through all request header providers in the registry.\n+ Additional context providers can be registered as described in\n+ :py:class:`mlflow.tracking.request_header.RequestHeaderProvider`.\n+\n+ Args:\n+ request_headers: A dictionary of request headers to override. If specified, headers passed\n+ in this argument will override those inferred from the context.\n+\n+ Returns:\n+ A dictionary of resolved headers.\n+ \"\"\"\n+\n+ all_request_headers = {}\n+ for provider in _request_header_provider_registry:\n+ try:\n+ if provider.in_context():\n+ # all_request_headers.update(provider.request_headers())\n+ for header, value in provider.request_headers().items():\n+ all_request_headers[header] = (\n+ f\"{all_request_headers[header]} {value}\"\n+ if header in all_request_headers\n+ else value\n+ )\n+ except Exception as e:\n+ _logger.warning(\"Encountered unexpected error during resolving request headers: %s\", e)\n+\n+ if request_headers is not None:\n+ all_request_headers.update(request_headers)\n+\n+ return all_request_headers\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/utils/databricks_utils.py b/mlflow/utils/databricks_utils.py\nindex 9d5d805ce..4d359d3e1 100644\n+++ a/mlflow/utils/databricks_utils.py\n--- b/mlflow/utils/databricks_utils.py\n@@ -32,7 +32,7 @@ from mlflow.legacy_databricks_cli.configure.provider import (\n )\n from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n+from mlflow.utils.rest_utils import MlflowHostCreds, http_request\n-from mlflow.utils.rest_utils import MlflowHostCreds\n from mlflow.utils.uri import (\n _DATABRICKS_UNITY_CATALOG_SCHEME,\n get_db_info_from_uri,\ndiff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..95d273bfa 100644\n+++ a/mlflow/utils/rest_utils.py\n--- b/mlflow/utils/rest_utils.py\n@@ -39,7 +39,7 @@ from mlflow.utils.request_utils import (\n augmented_raise_for_status, # noqa: F401\n cloud_storage_http_request, # noqa: F401\n )\n+from mlflow.utils.string_utils import strip_suffix\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -57,233 +57,233 @@ _DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING = (\n )\n \n \n+def http_request(\n+ host_creds,\n+ endpoint,\n+ method,\n+ max_retries=None,\n+ backoff_factor=None,\n+ backoff_jitter=None,\n+ extra_headers=None,\n+ retry_codes=_TRANSIENT_FAILURE_RESPONSE_CODES,\n+ timeout=None,\n+ raise_on_status=True,\n+ respect_retry_after_header=None,\n+ retry_timeout_seconds=None,\n+ **kwargs,\n+):\n+ \"\"\"Makes an HTTP request with the specified method to the specified hostname/endpoint. Transient\n+ errors such as Rate-limited (429), service unavailable (503) and internal error (500) are\n+ retried with an exponential back off with backoff_factor * (1, 2, 4, ... seconds).\n+ The function parses the API response (assumed to be JSON) into a Python object and returns it.\n \n+ Args:\n+ host_creds: A :py:class:`mlflow.rest_utils.MlflowHostCreds` object containing\n+ hostname and optional authentication.\n+ endpoint: A string for service endpoint, e.g. \"/path/to/object\".\n+ method: A string indicating the method to use, e.g. \"GET\", \"POST\", \"PUT\".\n+ max_retries: Maximum number of retries before throwing an exception.\n+ backoff_factor: A time factor for exponential backoff. e.g. value 5 means the HTTP\n+ request will be retried with interval 5, 10, 20... seconds. A value of 0 turns off the\n+ exponential backoff.\n+ backoff_jitter: A random jitter to add to the backoff interval.\n+ extra_headers: A dict of HTTP header name-value pairs to be included in the request.\n+ retry_codes: A list of HTTP response error codes that qualifies for retry.\n+ timeout: Wait for timeout seconds for response from remote server for connect and\n+ read request.\n+ raise_on_status: Whether to raise an exception, or return a response, if status falls\n+ in retry_codes range and retries have been exhausted.\n+ respect_retry_after_header: Whether to respect Retry-After header on status codes defined\n+ as Retry.RETRY_AFTER_STATUS_CODES or not.\n+ retry_timeout_seconds: Timeout for retries. Only effective when using Databricks SDK.\n+ kwargs: Additional keyword arguments to pass to `requests.Session.request()`\n+\n+ Returns:\n+ requests.Response object.\n+ \"\"\"\n+ cleaned_hostname = strip_suffix(host_creds.host, \"/\")\n+ url = f\"{cleaned_hostname}{endpoint}\"\n+\n+ # Set defaults for retry parameters from environment variables if not specified\n+ max_retries = MLFLOW_HTTP_REQUEST_MAX_RETRIES.get() if max_retries is None else max_retries\n+ backoff_factor = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_FACTOR.get() if backoff_factor is None else backoff_factor\n+ )\n+ backoff_jitter = (\n+ MLFLOW_HTTP_REQUEST_BACKOFF_JITTER.get() if backoff_jitter is None else backoff_jitter\n+ )\n+\n+ from mlflow.tracking.request_header.registry import resolve_request_headers\n+\n+ headers = dict(**resolve_request_headers())\n+ if extra_headers:\n+ headers = dict(**headers, **extra_headers)\n+\n+ if traffic_id := _MLFLOW_DATABRICKS_TRAFFIC_ID.get():\n+ headers[\"x-databricks-traffic-id\"] = traffic_id\n+\n+ if host_creds.use_databricks_sdk:\n+ from databricks.sdk.errors import DatabricksError\n+\n+ ws_client = get_workspace_client(\n+ host_creds.use_secret_scope_token,\n+ host_creds.host,\n+ host_creds.token,\n+ host_creds.databricks_auth_profile,\n+ retry_timeout_seconds=retry_timeout_seconds,\n+ timeout=timeout,\n+ )\n \n+ def make_sdk_call():\n+ # Databricks SDK `APIClient.do` API is for making request using\n+ # HTTP\n+ # https://github.com/databricks/databricks-sdk-py/blob/a714146d9c155dd1e3567475be78623f72028ee0/databricks/sdk/core.py#L134\n+ # suppress the warning due to https://github.com/databricks/databricks-sdk-py/issues/963\n+ with warnings.catch_warnings():\n+ warnings.filterwarnings(\n+ \"ignore\", message=f\".*{_DATABRICKS_SDK_RETRY_AFTER_SECS_DEPRECATION_WARNING}.*\"\n+ )\n+ raw_response = ws_client.api_client.do(\n+ method=method,\n+ path=endpoint,\n+ headers=headers,\n+ raw=True,\n+ query=kwargs.get(\"params\"),\n+ body=kwargs.get(\"json\"),\n+ files=kwargs.get(\"files\"),\n+ data=kwargs.get(\"data\"),\n+ )\n+ return raw_response[\"contents\"]._response\n \n+ try:\n+ # We retry the SDK call with exponential backoff because the Databricks SDK default\n+ # retry behavior does not handle all transient errors that we want to retry, and it\n+ # does not support a customizable retry policy based on HTTP response status codes.\n+ # Note that, in uncommon cases (due to the limited set if HTTP status codes and\n+ # response strings that Databricks SDK retries on), the SDK may retry internally,\n+ # and MLflow may retry on top of that, leading to more retries than specified by\n+ # `max_retries`. This is acceptable, given the enforcement of an overall request\n+ # timeout via `retry_timeout_seconds`.\n+ #\n+ # TODO: Update transient error handling defaults in Databricks SDK to match standard\n+ # practices (e.g. retrying on 429, 500, 503, etc.), support custom retries in Databricks\n+ # SDK, and remove this custom retry wrapper from MLflow\n+ return _retry_databricks_sdk_call_with_exponential_backoff(\n+ call_func=make_sdk_call,\n+ retry_codes=retry_codes,\n+ retry_timeout_seconds=(\n+ retry_timeout_seconds\n+ if retry_timeout_seconds is not None\n+ else MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get()\n+ ),\n+ backoff_factor=backoff_factor,\n+ backoff_jitter=backoff_jitter,\n+ max_retries=max_retries,\n+ )\n+ except DatabricksError as e:\n+ response = requests.Response()\n+ response.url = url\n+ response.status_code = ERROR_CODE_TO_HTTP_STATUS.get(e.error_code, 500)\n+ response.reason = str(e)\n+ response.encoding = \"UTF-8\"\n+ response._content = json.dumps(\n+ {\n+ \"error_code\": e.error_code,\n+ \"message\": str(e),\n+ }\n+ ).encode(\"UTF-8\")\n+ return response\n+\n+ _validate_max_retries(max_retries)\n+ _validate_backoff_factor(backoff_factor)\n+ respect_retry_after_header = (\n+ MLFLOW_HTTP_RESPECT_RETRY_AFTER_HEADER.get()\n+ if respect_retry_after_header is None\n+ else respect_retry_after_header\n+ )\n+\n+ timeout = MLFLOW_HTTP_REQUEST_TIMEOUT.get() if timeout is None else timeout\n+ auth_str = None\n+ if host_creds.username and host_creds.password:\n+ basic_auth_str = f\"{host_creds.username}:{host_creds.password}\".encode()\n+ auth_str = \"Basic \" + base64.standard_b64encode(basic_auth_str).decode(\"utf-8\")\n+ elif host_creds.token:\n+ auth_str = f\"Bearer {host_creds.token}\"\n+ elif host_creds.client_secret:\n+ raise MlflowException(\n+ \"To use OAuth authentication, set environmental variable \"\n+ f\"'{MLFLOW_ENABLE_DB_SDK.name}' to true\",\n+ error_code=CUSTOMER_UNAUTHORIZED,\n+ )\n \n+ if auth_str:\n+ headers[\"Authorization\"] = auth_str\n \n+ if host_creds.client_cert_path is not None:\n+ kwargs[\"cert\"] = host_creds.client_cert_path\n \n+ if host_creds.aws_sigv4:\n+ # will overwrite the Authorization header\n+ from requests_auth_aws_sigv4 import AWSSigV4\n \n+ kwargs[\"auth\"] = AWSSigV4(\"execute-api\")\n+ elif host_creds.auth:\n+ from mlflow.tracking.request_auth.registry import fetch_auth\n \n+ kwargs[\"auth\"] = fetch_auth(host_creds.auth)\n \n+ try:\n+ return _get_http_response_with_retries(\n+ method,\n+ url,\n+ max_retries,\n+ backoff_factor,\n+ backoff_jitter,\n+ retry_codes,\n+ raise_on_status,\n+ headers=headers,\n+ verify=host_creds.verify,\n+ timeout=timeout,\n+ respect_retry_after_header=respect_retry_after_header,\n+ **kwargs,\n+ )\n+ except requests.exceptions.Timeout as to:\n+ raise MlflowException(\n+ f\"API request to {url} failed with timeout exception {to}.\"\n+ \" To increase the timeout, set the environment variable \"\n+ f\"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value.\"\n+ ) from to\n+ except requests.exceptions.InvalidURL as iu:\n+ raise InvalidUrlException(f\"Invalid url: {url}\") from iu\n+ except Exception as e:\n+ raise MlflowException(f\"API request to {url} failed with exception {e}\")\n+\n+\n+@lru_cache(maxsize=5)\n+def get_workspace_client(\n+ use_secret_scope_token,\n+ host,\n+ token,\n+ databricks_auth_profile,\n+ retry_timeout_seconds=None,\n+ timeout=None,\n+):\n+ from databricks.sdk import WorkspaceClient\n+ from databricks.sdk.config import Config\n \n+ if use_secret_scope_token:\n+ kwargs = {\"host\": host, \"token\": token}\n+ else:\n+ kwargs = {\"profile\": databricks_auth_profile}\n+ if timeout is not None:\n+ kwargs[\"http_timeout_seconds\"] = timeout\n+ config = Config(\n+ **kwargs,\n+ retry_timeout_seconds=retry_timeout_seconds\n+ or MLFLOW_DATABRICKS_ENDPOINT_HTTP_RETRY_TIMEOUT.get(),\n+ )\n+ # Note: If we use `config` param, all SDK configurations must be set in `config` object.\n+ return WorkspaceClient(config=config)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _can_parse_as_json_object(string):\n@@ -384,25 +384,25 @@ def _validate_backoff_factor(backoff_factor):\n )\n \n \n+def validate_deployment_timeout_config(timeout: int | None, retry_timeout_seconds: int | None):\n+ \"\"\"\n+ Validate that total retry timeout is not less than single request timeout.\n \n+ Args:\n+ timeout: Maximum time for a single HTTP request (in seconds)\n+ retry_timeout_seconds: Maximum time for all retry attempts combined (in seconds)\n+ \"\"\"\n+ if timeout is not None and retry_timeout_seconds is not None:\n+ if retry_timeout_seconds < timeout:\n+ warnings.warn(\n+ f\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT ({retry_timeout_seconds}s) is set \"\n+ f\"lower than MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT ({timeout}s). This means the \"\n+ \"total retry timeout could expire before a single request completes, causing \"\n+ \"premature failures. For long-running predictions, ensure \"\n+ \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT >= MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT. \"\n+ f\"Recommended: Set MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least {timeout}s.\",\n+ stacklevel=2,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _time_sleep(seconds: float) -> None:\n@@ -414,78 +414,78 @@ def _time_sleep(seconds: float) -> None:\n time.sleep(seconds)\n \n \n+def _retry_databricks_sdk_call_with_exponential_backoff(\n+ *,\n+ call_func: Callable[..., Any],\n+ retry_codes: list[int],\n+ retry_timeout_seconds: int,\n+ backoff_factor: int,\n+ backoff_jitter: float,\n+ max_retries: int,\n+):\n+ \"\"\"\n+ Retry a Databricks SDK call with exponential backoff until timeout or max retries reached.\n \n+ Args:\n+ call_func: Function to call that may raise DatabricksError\n+ retry_codes: Set of HTTP status codes that should trigger retries\n+ retry_timeout_seconds: Maximum time to spend retrying in seconds\n+ backoff_factor: Factor for exponential backoff\n+ backoff_jitter: Random jitter to add to backoff\n+ max_retries: Maximum number of retry attempts\n+\n+ Returns:\n+ The result of call_func() on success\n+\n+ Raises:\n+ DatabricksError: If all retries are exhausted or non-retryable error occurs\n+ \"\"\"\n+ from databricks.sdk.errors import STATUS_CODE_MAPPING, DatabricksError\n \n+ start_time = time.time()\n+ attempt = 0\n \n+ while attempt <= max_retries:\n+ try:\n+ return call_func()\n+ except DatabricksError as e:\n+ # Get HTTP status code from the error\n+ status_code = next(\n+ (code for code, cls in STATUS_CODE_MAPPING.items() if isinstance(e, cls)), 500\n+ )\n+ # Check if this is a retryable error\n+ if status_code not in retry_codes:\n+ raise\n+\n+ # Check if we've exceeded max retries\n+ if attempt >= max_retries:\n+ _logger.warning(f\"Max retries ({max_retries}) exceeded: {e}\")\n+ raise\n+\n+ # Calculate backoff time with exponential backoff and jitter\n+ # NB: Ideally, we'd use urllib3.Retry to compute the jitter, check whether we've\n+ # exceed max retries, etc. However, urllib3.Retry in urllib3 version 1.x, which MLflow\n+ # maintains compatibility with, doesn't support retries with jitter\n+ if attempt <= 0:\n+ backoff_time = 0 # No backoff on first retry attempt\n+ else:\n+ backoff_time = backoff_factor * (2**attempt)\n+ if backoff_jitter > 0:\n+ backoff_time += random.random() * backoff_jitter\n+\n+ # Check if we've exceeded or would exceed timeout\n+ elapsed_time = time.time() - start_time\n+ if elapsed_time + backoff_time >= retry_timeout_seconds:\n+ _logger.warning(f\"Retry timeout ({retry_timeout_seconds}s) exceeded: {e}\")\n+ raise\n+\n+ _logger.debug(\n+ f\"Databricks SDK call failed with retryable error (status {status_code}): {e}. \"\n+ f\"Retrying in {backoff_time:.2f} seconds (attempt {attempt + 1})\"\n+ )\n \n+ _time_sleep(backoff_time)\n+ attempt += 1\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _get_path(path_prefix, endpoint_path):\n@@ -720,4 +720,4 @@ class MlflowHostCreds:\n if self.server_cert_path is None:\n return not self.ignore_tls_verification\n else:\n+ return self.server_cert_path\n- return self.server_cert_path\n\\ No newline at end of file\ndiff --git a/mlflow/utils/string_utils.py b/mlflow/utils/string_utils.py\nindex 9628a782b..46a49e8fc 100644\n+++ a/mlflow/utils/string_utils.py\n--- b/mlflow/utils/string_utils.py\n@@ -12,10 +12,10 @@ def strip_prefix(original: str, prefix: str) -> str:\n return original\n \n \n+def strip_suffix(original: str, suffix: str) -> str:\n+ if original.endswith(suffix) and suffix != \"\":\n+ return original[: -len(suffix)]\n+ return original\n-\n-\n-\n-\n \n \n def is_string_type(item: Any) -> bool:\n@@ -188,4 +188,4 @@ def format_table_cell_value(field: str, cell_value: Any, values: list[Any] | Non\n if len(str(cell_value)) > 20:\n cell_value = str(cell_value)[:17] + \"...\"\n \n+ return str(cell_value)\n- return str(cell_value)\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py\nindex 368143ad8..64d50e488 100644\n--- a/mlflow/utils/rest_utils.py\n+++ b/mlflow/utils/rest_utils.py\n@@ -259,7 +259,6 @@ def http_request(\n raise MlflowException(f\"API request to {url} failed with exception {e}\")\n \n \n-@lru_cache(maxsize=5)\n def get_workspace_client(\n use_secret_scope_token,\n host,\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py\ndeleted file mode 100644\nindex db1f15091..000000000\n--- a/tests/deployments/databricks/test_databricks.py\n+++ /dev/null\n@@ -1,605 +0,0 @@\n-import os\n-import warnings\n-from unittest import mock\n-\n-import pytest\n-\n-from mlflow.deployments import get_deploy_client\n-from mlflow.exceptions import MlflowException\n-\n-\n-@pytest.fixture(autouse=True)\n-def mock_databricks_credentials(monkeypatch):\n- monkeypatch.setenv(\"DATABRICKS_HOST\", \"https://test.cloud.databricks.com\")\n- monkeypatch.setenv(\"DATABRICKS_TOKEN\", \"secret\")\n-\n-\n-def test_get_deploy_client():\n- get_deploy_client(\"databricks\")\n- get_deploy_client(\"databricks://scope:prefix\")\n-\n-\n-def test_create_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_config_only():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.create_endpoint(\n- config={\n- \"name\": \"test_new\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test_entity\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"route_optimized\": True,\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_match():\n- \"\"\"Test when name is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning about using name parameter.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name' as a parameter is deprecated. \"\n- \"Please specify 'name' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"name\": \"test\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_name_mismatch():\n- \"\"\"Test when name is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Name mismatch. Found 'test1' as parameter and 'test2' \"\n- \"in config. Please specify 'name' only within the config \"\n- \"dictionary as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test1\",\n- config={\n- \"name\": \"test2\",\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_route_optimized_match():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with matching values.\n- Should emit a deprecation warning.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'route_optimized' as a parameter is deprecated. \"\n- \"Please specify 'route_optimized' only within the config dictionary.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": True,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_route_optimized_mismatch():\n- \"\"\"Test when route_optimized is provided both in config and as named arg with different values.\n- Should raise an MlflowException.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.raises(\n- MlflowException,\n- match=\"Conflicting 'route_optimized' values found. \"\n- \"Please specify 'route_optimized' only within the config dictionary \"\n- \"as this parameter is deprecated.\",\n- ):\n- client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"name\": \"test\",\n- \"route_optimized\": False,\n- \"config\": {\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- },\n- )\n- mock_request.assert_not_called()\n-\n-\n-def test_create_endpoint_named_name():\n- \"\"\"Test using the legacy format with separate parameters instead of full API payload.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_create_endpoint_named_route_optimized():\n- \"\"\"Test using the legacy format with route_optimized parameter.\n- Should emit a deprecation warning about the old format.\n- \"\"\"\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"Passing 'name', 'config', and 'route_optimized' as separate parameters is \"\n- \"deprecated. Please pass the full API request payload as a single dictionary \"\n- \"in the 'config' parameter.\",\n- ):\n- resp = client.create_endpoint(\n- name=\"test\",\n- route_optimized=True,\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_get_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"name\": \"test\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.get_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {\"name\": \"test\"}\n-\n-\n-def test_list_endpoints():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"endpoints\": [{\"name\": \"test\"}]}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.list_endpoints()\n- mock_request.assert_called_once()\n- assert resp == [{\"name\": \"test\"}]\n-\n-\n-def test_update_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- with pytest.warns(\n- UserWarning,\n- match=\"The `update_endpoint` method is deprecated. Use the specific update methods—\"\n- \"`update_endpoint_config`, `update_endpoint_tags`, `update_endpoint_rate_limits`, \"\n- \"`update_endpoint_ai_gateway`—instead.\",\n- ):\n- resp = client.update_endpoint(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"test\",\n- \"external_model\": {\n- \"name\": \"gpt-4\",\n- \"provider\": \"openai\",\n- \"openai_config\": {\n- \"openai_api_key\": \"secret\",\n- },\n- },\n- }\n- ],\n- \"task\": \"llm/v1/chat\",\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_config():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_config(\n- endpoint=\"test\",\n- config={\n- \"served_entities\": [\n- {\n- \"name\": \"gpt-4-mini\",\n- \"external_model\": {\n- \"name\": \"gpt-4-mini\",\n- \"provider\": \"openai\",\n- \"task\": \"llm/v1/chat\",\n- \"openai_config\": {\n- \"openai_api_key\": \"{{secrets/scope/key}}\",\n- },\n- },\n- }\n- ],\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_tags():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_tags(\n- endpoint=\"test\",\n- config={\"add_tags\": [{\"key\": \"project\", \"value\": \"test\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_rate_limits():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_rate_limits(\n- endpoint=\"test\",\n- config={\"rate_limits\": [{\"calls\": 10, \"key\": \"endpoint\", \"renewal_period\": \"minute\"}]},\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_update_endpoint_ai_gateway():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.update_endpoint_ai_gateway(\n- endpoint=\"test\",\n- config={\n- \"usage_tracking_config\": {\"enabled\": True},\n- \"inference_table_config\": {\n- \"enabled\": True,\n- \"catalog_name\": \"my_catalog\",\n- \"schema_name\": \"my_schema\",\n- },\n- },\n- )\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_delete_endpoint():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.delete_endpoint(endpoint=\"test\")\n- mock_request.assert_called_once()\n- assert resp == {}\n-\n-\n-def test_predict():\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- with mock.patch(\"requests.Session.request\", return_value=mock_resp) as mock_request:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_request.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert resp == {\"foo\": \"bar\"}\n-\n-\n-def test_predict_stream_with_total_timeout_env_var(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"900\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n- mock_http.assert_called_once()\n- call_kwargs = mock_http.call_args[1]\n- assert call_kwargs[\"retry_timeout_seconds\"] == 900\n- assert len(chunks) == 1\n-\n-\n-def test_predict_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- resp = client.predict(endpoint=\"test\", inputs={})\n-\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_stream_warns_on_misconfigured_timeouts(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"300\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"120\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.iter_lines.return_value = [\n- \"data: \" + '{\"id\": \"1\", \"choices\": [{\"delta\": {\"content\": \"Hello\"}}]}',\n- \"data: [DONE]\",\n- ]\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n- mock_resp.encoding = \"utf-8\"\n-\n- with mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http:\n- with warnings.catch_warnings(record=True) as w:\n- warnings.simplefilter(\"always\")\n- chunks = list(client.predict_stream(endpoint=\"test\", inputs={}))\n-\n- mock_http.assert_called_once()\n- assert len(chunks) == 1\n- assert len(w) == 1\n- warning_msg = str(w[0].message)\n- assert \"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\" in warning_msg\n- assert \"(120s)\" in warning_msg\n- assert \"(300s)\" in warning_msg\n-\n-\n-def test_predict_no_warning_when_timeouts_properly_configured(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT\", \"120\")\n- monkeypatch.setenv(\"MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT\", \"600\")\n- client = get_deploy_client(\"databricks\")\n- mock_resp = mock.Mock()\n- mock_resp.json.return_value = {\"foo\": \"bar\"}\n- mock_resp.url = os.environ[\"DATABRICKS_HOST\"]\n- mock_resp.status_code = 200\n-\n- with (\n- mock.patch(\n- \"mlflow.deployments.databricks.http_request\", return_value=mock_resp\n- ) as mock_http,\n- mock.patch(\"mlflow.utils.rest_utils._logger.warning\") as mock_warning,\n- ):\n- resp = client.predict(endpoint=\"test\", inputs={})\n- mock_http.assert_called_once()\n- assert resp == {\"foo\": \"bar\"}\n- mock_warning.assert_not_called()\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S26\",\n \"shared_functions\": 0,\n \"shared_lines\": 4\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S11\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":73,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S1\",\n \"text\": \"Extract score and justification from LLM judge model response text. This function attempts to parse a score (integer) and justification (string) from the response text of an LLM judge model. It supports both JSON format and plain text format with specific patterns. Returns: tuple[int | None, str | None]: A tuple containing: - score (int | None): The extracted numerical score as an integer, or None if extraction failed - justification (str | None): The extracted justification text, or None if the input text is empty/None. If extraction fails, returns an error message containing the raw output. Robust parsing of LLM judge responses to extract numerical scores and text justifications\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"78-105\",\n \"function\": \"_extract_score_and_justification\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..b82de8dcc 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -88,7 +88,7 @@ def _extract_score_and_justification(text):\n if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n ):\n- score = int(match.group(1))\n+ score = float(match.group(1))\n justification = match.group(2)\n else:\n score = None\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S4\",\n \"shared_functions\": 1,\n \"shared_lines\": 27\n },\n {\n \"spec_id\": \"S3\",\n \"shared_functions\": 1,\n \"shared_lines\": 9\n },\n {\n \"spec_id\": \"S6\",\n \"shared_functions\": 1,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S2\",\n \"shared_functions\": 1,\n \"shared_lines\": 5\n },\n {\n \"spec_id\": \"S5\",\n \"shared_functions\": 1,\n \"shared_lines\": 3\n },\n {\n \"spec_id\": \"S20\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":74,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S5\",\n \"text\": \"The function first normalizes the text by converting \\\"score\\\" and \\\"justification\\\" keywords to lowercase using case-insensitive regex replacement\",\n \"category\": \"implementation_constraint\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"80-82\",\n \"function\": \"_extract_score_and_justification\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..771dea1bd 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -76,7 +76,7 @@ def _format_args_string(grading_context_columns: list[str] | None, eval_values,\n # Function to extract Score and Justification\n def _extract_score_and_justification(text):\n if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"[Ss]core\", \"score\", text)\n text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n # Attempt to parse JSON\n try:\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 1,\n \"shared_lines\": 3\n },\n {\n \"spec_id\": \"S4\",\n \"shared_functions\": 1,\n \"shared_lines\": 3\n },\n {\n \"spec_id\": \"S2\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":75,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S6\",\n \"text\": \"If the extracted score is not a number or justification is not a string, returns None for score and an error message for justification. If extraction fails, returns an error message containing the raw output.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"97-103\",\n \"function\": \"_extract_score_and_justification\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..d2b277890 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -95,6 +95,8 @@ def _extract_score_and_justification(text):\n justification = f\"Failed to extract score and justification. Raw output: {text}\"\n \n if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ if isinstance(score, (int, float)) and isinstance(justification, (int, float)):\n+ return score, str(justification)\n return None, f\"Failed to extract score and justification. Raw output: {text}\"\n \n return score, justification\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 1,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S4\",\n \"shared_functions\": 1,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S2\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":76,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S7\",\n \"text\": \"def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str: Format grading context columns into a structured string for LLM evaluation prompts. This function extracts values from specified grading context columns and formats them into a human-readable string that can be included in evaluation prompts for LLM-as-a-judge metrics. Parameters: grading_context_columns (list[str] | None): List of column names to extract values from. eval_values (dict): Dictionary containing the evaluation data, where keys are column names and values are either pandas Series or list-like objects containing the actual data. indx (int): Index position to extract the specific value from each column's data series. Returns: str: A formatted string containing the grading context information. Otherwise, returns a multi-line string with the format: \\\"Additional information used by the model:\\\\nkey: {column_name}\\\\nvalue:\\\\n{column_value}\\\\n...\\\" Each grading context item is formatted with a \\\"key:\\\" and \\\"value:\\\" structure\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"48-74\",\n \"function\": \"_format_args_string\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..23b88abe3 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -67,7 +67,10 @@ def _format_args_string(grading_context_columns: list[str] | None, eval_values,\n else (\n \"Additional information used by the model:\\n\"\n + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ [\n+ f\"key: {arg.replace('_', ' ')}\\nvalue:\\n{arg_value}\"\n+ for arg, arg_value in args_dict.items()\n+ ]\n )\n )\n )\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S8\",\n \"shared_functions\": 1,\n \"shared_lines\": 11\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 1,\n \"shared_lines\": 6\n },\n {\n \"spec_id\": \"S9\",\n \"shared_functions\": 1,\n \"shared_lines\": 4\n }\n ]\n}"}}},{"rowIdx":77,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S8\",\n \"text\": \"Returns an empty string if grading_context_columns is None or empty. Empty or None grading_context_columns will result in an empty string return\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"64-74\",\n \"function\": \"_format_args_string\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c96cb7d2e 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -63,7 +63,7 @@ def _format_args_string(grading_context_columns: list[str] | None, eval_values,\n \n return (\n \"\"\n- if args_dict is None or len(args_dict) == 0\n+ if args_dict is None or len(eval_values) == 0\n else (\n \"Additional information used by the model:\\n\"\n + \"\\n\".join(\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S7\",\n \"shared_functions\": 1,\n \"shared_lines\": 11\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S9\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":78,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S9\",\n \"text\": \"Raises: MlflowException: If any column specified in grading_context_columns does not exist in the eval_values dictionary. The exception message will indicate which column is missing and list all available columns.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"59-62\",\n \"function\": \"_format_args_string\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..5097476bb 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -56,7 +56,7 @@ def _format_args_string(grading_context_columns: list[str] | None, eval_values,\n if isinstance(eval_values[arg], pd.Series)\n else eval_values[arg][indx]\n )\n- else:\n+ elif args_dict:\n raise MlflowException(\n f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n )\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S7\",\n \"shared_functions\": 1,\n \"shared_lines\": 4\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S8\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":79,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S11\",\n \"text\": \"def make_genai_metric(name: str, definition: str, grading_prompt: str, examples: list[EvaluationExample] | None = None, version: str | None = _get_latest_metric_version(), model: str | None = _get_default_model(), grading_context_columns: str | list[str] | None = None, include_input: bool = True, parameters: dict[str, Any] | None = None, aggregations: list[str] | None = None, greater_is_better: bool = True, max_workers: int = 10, metric_metadata: dict[str, Any] | None = None, extra_headers: dict[str, str] | None = None, proxy_url: str | None = None) -> EvaluationMetric: Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. Returns: A metric object. Args: name: Name of the metric. definition: Definition of the metric. grading_prompt: Grading criteria of the metric. examples: (Optional) Examples of the metric. version: (Optional) Version of the metric. model: (Optional) Model uri of the judge model that will be used to compute the metric. greater_is_better: (Optional) Whether the metric is better when it is greater. metric_metadata: (Optional) Dictionary of metadata to be attached to the EvaluationMetric object. Support both pre-built and custom metric creation with configurable grading criteria\",\n \"category\": \"structural\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"398-556\",\n \"function\": \"make_genai_metric\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..eb1606e76 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -371,7 +371,7 @@ def make_genai_metric(\n include_input: bool = True,\n parameters: dict[str, Any] | None = None,\n aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n+ greater_is_better: bool = False,\n max_workers: int = 10,\n metric_metadata: dict[str, Any] | None = None,\n extra_headers: dict[str, str] | None = None,\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S13\",\n \"shared_functions\": 1,\n \"shared_lines\": 29\n },\n {\n \"spec_id\": \"S15\",\n \"shared_functions\": 1,\n \"shared_lines\": 21\n },\n {\n \"spec_id\": \"S14\",\n \"shared_functions\": 1,\n \"shared_lines\": 16\n },\n {\n \"spec_id\": \"S12\",\n \"shared_functions\": 1,\n \"shared_lines\": 10\n },\n {\n \"spec_id\": \"S19\",\n \"shared_functions\": 1,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S18\",\n \"shared_functions\": 1,\n \"shared_lines\": 4\n },\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S16\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":80,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S12\",\n \"text\": \"The full grading prompt is stored in the metric_details field of the ``EvaluationMetric`` object. metric_details: (Optional) A description of the metric and how it is calculated.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"547-556\",\n \"function\": \"make_genai_metric\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..4885a4619 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -665,12 +665,16 @@ def make_genai_metric(\n # extra params in grading_context_columns can only be passed as positional args\n eval_fn.__signature__ = Signature(signature_parameters)\n \n+ metric_details = evaluation_context[\"eval_prompt\"].__str__()\n+ if not include_input:\n+ metric_details = metric_details.split(\"\\nMetric definition:\")[0]\n+\n return _make_metric(\n eval_fn=eval_fn,\n greater_is_better=greater_is_better,\n name=name,\n version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_details=metric_details,\n metric_metadata=metric_metadata,\n genai_metric_args=genai_metric_args,\n require_strict_signature=True,\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S11\",\n \"shared_functions\": 1,\n \"shared_lines\": 10\n },\n {\n \"spec_id\": \"S19\",\n \"shared_functions\": 1,\n \"shared_lines\": 6\n },\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S13\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S14\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S15\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S16\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S18\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":81,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S13\",\n \"text\": \"Provide example-based few-shot learning for evaluation consistency. Also raised when grading context columns are malformed or when required columns are missing from example grading context.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"500-528\",\n \"function\": \"make_genai_metric\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..b2310a624 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -514,9 +514,12 @@ def make_genai_metric(\n elif isinstance(example.grading_context, dict):\n grading_context = example.grading_context\n else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n+ # The grading context is string-like. Assume that it corresponds to the\n+ # grading context columns and update the example accordingly\n+ grading_context = {\n+ column: example.grading_context\n+ for column in (grading_context_columns or [grading_context_columns[0]])\n+ }\n example.grading_context = grading_context\n \n if set(grading_context.keys()) != set(grading_context_columns):\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S11\",\n \"shared_functions\": 1,\n \"shared_lines\": 29\n },\n {\n \"spec_id\": \"S18\",\n \"shared_functions\": 1,\n \"shared_lines\": 4\n },\n {\n \"spec_id\": \"S12\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S14\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S15\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S16\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S19\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":82,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S14\",\n \"text\": \"Raises: MlflowException: If the specified version is not supported or if there are issues with the evaluation model construction.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"530-545\",\n \"function\": \"make_genai_metric\"\n },\n {\n \"file\": \"mlflow/utils/class_utils.py\",\n \"lines\": \"4-6\",\n \"function\": \"_get_class_from_string\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..07c6f18ee 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -539,7 +539,9 @@ def make_genai_metric(\n if examples is not None:\n examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ class_name = (\n+ f\"mlflow.metrics.genai.prompts.{version or _get_latest_metric_version()}.EvaluationModel\"\n+ )\n try:\n evaluation_model_class_module = _get_class_from_string(class_name)\n except ModuleNotFoundError:\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S11\",\n \"shared_functions\": 1,\n \"shared_lines\": 16\n },\n {\n \"spec_id\": \"S12\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S13\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":83,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S15\",\n \"text\": \"Backward compatibility for metric serialization/deserialization across different MLflow versions. The metric configuration is serialized and stored as an artifact to enable later deserialization and result interpretation. genai_metric_args: (Optional) A dictionary containing arguments specified by users when calling make_genai_metric or make_genai_metric_from_prompt. Those args are persisted so that we can deserialize the same metric object later.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"479-499\",\n \"function\": \"make_genai_metric\"\n },\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"271-289\",\n \"function\": \"make_genai_metric_from_prompt\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..8a776aab8 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -496,7 +496,7 @@ def make_genai_metric(\n \"aggregations\": aggregations,\n \"greater_is_better\": greater_is_better,\n \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n+ \"metric_metadata\": None,\n # Record the mlflow version for serialization in case the function signature changes later\n \"mlflow_version\": VERSION,\n \"fn_name\": make_genai_metric.__name__,\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S11\",\n \"shared_functions\": 1,\n \"shared_lines\": 21\n },\n {\n \"spec_id\": \"S12\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S13\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S20\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S23\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":84,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S17\",\n \"text\": \"aggregations: (Optional) The list of options to aggregate the scores. Currently supported options are: min, max, mean, median, variance, p90. Aggregate evaluation results across multiple data points. Generate structured evaluation outputs with per-row scores, justifications, and aggregate statistics. Returns: MetricValue with per-row scores, per-row justifications, and aggregate results.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"196-225\",\n \"function\": \"_get_aggregate_results\"\n },\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"634-636\",\n \"function\": \"make_genai_metric.eval_fn\"\n },\n {\n \"file\": \"mlflow/metrics/base.py\",\n \"lines\": \"32-38\",\n \"function\": \"MetricValue.__post_init__\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..35a504924 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -177,7 +177,7 @@ def _get_aggregate_results(scores, aggregations):\n \"min\": np.min,\n \"max\": np.max,\n \"mean\": np.mean,\n- \"median\": np.median,\n+ \"median\": np.mean,\n \"variance\": np.var,\n \"p90\": lambda x: np.percentile(x, 90) if x else None,\n }\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S18\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S21\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S22\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S32\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S33\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":85,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S19\",\n \"text\": \"parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"546-552\",\n \"function\": \"make_genai_metric\"\n },\n {\n \"file\": \"mlflow/metrics/genai/prompts/v1.py\",\n \"lines\": \"108-110\",\n \"function\": \"EvaluationModel\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..7f88b16b4 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -125,7 +125,7 @@ class EvaluationModel:\n grading_prompt=self.grading_prompt,\n examples=examples_str,\n ),\n- \"parameters\": self.parameters,\n+ \"parameters\": {**self.parameters, **default_parameters},\n }\n \n def _format_examples(self):\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S11\",\n \"shared_functions\": 1,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S12\",\n \"shared_functions\": 1,\n \"shared_lines\": 6\n },\n {\n \"spec_id\": \"S13\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S26\",\n \"shared_functions\": 0,\n \"shared_lines\": 3\n }\n ]\n}"}}},{"rowIdx":86,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S20\",\n \"text\": \"def make_genai_metric_from_prompt(name: str, judge_prompt: str | None = None, model: str | None = _get_default_model(), parameters: dict[str, Any] | None = None, aggregations: list[str] | None = None, greater_is_better: bool = True, max_workers: int = 10, metric_metadata: dict[str, Any] | None = None, extra_headers: dict[str, str] | None = None, proxy_url: str | None = None) -> EvaluationMetric: This produces a metric using only the supplied judge prompt without any pre-written system prompt. judge_prompt: The entire prompt to be used for the judge model. The prompt will be minimally wrapped in formatting instructions to ensure scores can be parsed. The judge prompt will be automatically wrapped with formatting instructions to ensure proper score parsing. Returns: A metric object.\",\n \"category\": \"structural\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"229-241\",\n \"function\": \"make_genai_metric_from_prompt\"\n },\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"267-269\",\n \"function\": \"make_genai_metric_from_prompt\"\n },\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"348-355\",\n \"function\": \"make_genai_metric_from_prompt\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..41254bba9 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -288,7 +288,15 @@ def make_genai_metric_from_prompt(\n \"\"\"\n import numpy as np\n \n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ _judge_system_prompt = (\n+ \"You are an impartial judge. You will be given an output that a machine learning \"\n+ \"model produced. Evaluate it carefully according to the instructions below.\\n\\n\"\n+ )\n+ prompt_template = PromptTemplate(\n+ [judge_prompt, _PROMPT_FORMATTING_WRAPPER]\n+ if parameters is not None\n+ else [_judge_system_prompt, judge_prompt, _PROMPT_FORMATTING_WRAPPER]\n+ )\n allowed_variables = prompt_template.variables\n additional_variables = list(allowed_variables - {\"predictions\"})\n \n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S23\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S15\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":87,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S21\",\n \"text\": \"include_input: (Optional) Whether to include the input when computing the metric.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"515-525\",\n \"function\": \"make_genai_metric.process_example\"\n },\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"605-612\",\n \"function\": \"make_genai_metric.eval_fn\"\n },\n {\n \"file\": \"mlflow/metrics/genai/prompts/v1.py\",\n \"lines\": \"116-118\",\n \"function\": \"EvaluationModel.to_dict\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..b36efd754 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -526,7 +526,7 @@ def make_genai_metric(\n f\" Required grading context columns: {grading_context_columns}\\n\"\n )\n \n- if not include_input:\n+ if not include_input and grading_context_columns:\n return EvaluationExample(\n output=example.output,\n score=example.score,\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S26\",\n \"shared_functions\": 1,\n \"shared_lines\": 3\n },\n {\n \"spec_id\": \"S17\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S18\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":88,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S23\",\n \"text\": \"The prompt may use f-string formatting to include variables. Corresponding variables must be passed as keyword arguments into the resulting metric's eval function. Variables in the judge prompt using f-string formatting must be provided as arguments to the metric's eval function. Raises: MlflowException: If there are issues with model scoring, invalid parameters, or missing required variables.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"292-330\",\n \"function\": \"make_genai_metric_from_prompt.eval_fn\"\n },\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"268-270\",\n \"function\": \"make_genai_metric_from_prompt\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..dc52343cb 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -324,7 +324,7 @@ def make_genai_metric_from_prompt(\n kwargs[\"predictions\"] = arg\n else:\n kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n+ if kwargs and (missing_variables := allowed_variables - set(kwargs.keys())):\n raise MlflowException(\n message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n error_code=INVALID_PARAMETER_VALUE,\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S20\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S15\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S32\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":89,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S29\",\n \"text\": \"def __str__(self): Return a string representation of the EvaluationMetric instance. Returns: str: A formatted string representation of the EvaluationMetric instance in the format \\\"EvaluationMetric(name=<name>, greater_is_better=<boolean>, ...)\\\" where additional attributes like long_name, version, metric_details, and metric_metadata are included only if they have non-None values.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/models/evaluation/base.py\",\n \"lines\": \"146-158\",\n \"function\": \"EvaluationMetric.__str__\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..41c160b66 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -148,8 +148,7 @@ class EvaluationMetric:\n \n if self.long_name:\n parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n+ parts.append(f\"version={self.version}\")\n if self.metric_details:\n parts.append(f\"metric_details={self.metric_details}\")\n if self.metric_metadata:\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S30\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":90,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S30\",\n \"text\": \"Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric on the given input data. The generated class is callable with a __call__ method that takes the arguments specified in the signature of the eval_fn function. Args: eval_fn: the evaluation function of the EvaluationMetric. Returns: A dynamically generated callable CallableEvaluationMetric class.\",\n \"category\": \"structural\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/models/evaluation/base.py\",\n \"lines\": \"163-330\",\n \"function\": \"_generate_eval_metric_class\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..984160a9c 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -275,7 +275,11 @@ def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n ) -> MetricValue:\n return self.eval_fn(**kwargs)\n \n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ allowed_kwargs_params = {\n+ name: param\n+ for name, param in inspect.signature(eval_fn).parameters.items()\n+ if param.kind is not Parameter.POSITIONAL_OR_KEYWORD\n+ }\n _call_method.__signature__ = Signature(\n parameters=[\n Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S31\",\n \"shared_functions\": 1,\n \"shared_lines\": 77\n },\n {\n \"spec_id\": \"S10\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S29\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":91,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S32\",\n \"text\": \"extra_headers: (Optional) Additional headers to be passed to the judge model.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"620-626\",\n \"function\": \"make_genai_metric.eval_fn\"\n },\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"331-333\",\n \"function\": \"make_genai_metric_from_prompt.eval_fn\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..d577e0e11 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -333,7 +333,7 @@ def make_genai_metric_from_prompt(\n grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ arg_strings, model, parameters, None, proxy_url, max_workers\n )\n \n aggregate_scores = _get_aggregate_results(scores, aggregations)\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S33\",\n \"shared_functions\": 2,\n \"shared_lines\": 10\n },\n {\n \"spec_id\": \"S22\",\n \"shared_functions\": 1,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S17\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S23\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":92,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S33\",\n \"text\": \"proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the judge model is served via a proxy endpoint, not directly via LLM provider services. If not specified, the default URL for the LLM provider will be used (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"620-626\",\n \"function\": \"make_genai_metric.eval_fn\"\n },\n {\n \"file\": \"mlflow/metrics/genai/genai_metric.py\",\n \"lines\": \"331-333\",\n \"function\": \"make_genai_metric_from_prompt.eval_fn\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n--- a/mlflow/metrics/__init__.py\n+++ b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n-from mlflow.models import (\n- EvaluationMetric,\n- make_metric,\n-)\n+from mlflow.models import EvaluationMetric\n+\n+\n+\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n--- a/mlflow/metrics/base.py\n+++ b/mlflow/metrics/base.py\n@@ -29,10 +29,8 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n- def __post_init__(self):\n- if (\n- self.aggregate_results is None\n- and isinstance(self.scores, (list, tuple))\n- and all(_is_numeric(score) for score in self.scores)\n- ):\n- self.aggregate_results = standard_aggregations(self.scores)\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n--- a/mlflow/metrics/genai/__init__.py\n+++ b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import (\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n- retrieve_custom_metrics,\n-)\n+from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n--- a/mlflow/metrics/genai/base.py\n+++ b/mlflow/metrics/genai/base.py\n@@ -63,38 +63,35 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n- def _format_grading_context(self):\n- if isinstance(self.grading_context, dict):\n- return \"\\n\".join(\n- [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n- )\n- else:\n- return self.grading_context\n-\n- def __str__(self) -> str:\n- return PromptTemplate(\n- [\n- \"\"\"\n-Example Input:\n-{input}\n-\"\"\",\n- \"\"\"\n-Example Output:\n-{output}\n-\"\"\",\n- \"\"\"\n-Additional information used by the model:\n-{grading_context}\n-\"\"\",\n- \"\"\"\n-Example score: {score}\n-Example justification: {justification}\n- \"\"\",\n- ]\n- ).format(\n- input=self.input,\n- output=self.output,\n- grading_context=self._format_grading_context(),\n- score=self.score,\n- justification=self.justification,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n-from mlflow.models import EvaluationMetric, make_metric\n-from mlflow.models.evaluation.base import _make_metric\n+from mlflow.models import EvaluationMetric\n+\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n-def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n- import pandas as pd\n \n- args_dict = {}\n- for arg in grading_context_columns:\n- if arg in eval_values:\n- args_dict[arg] = (\n- eval_values[arg].iloc[indx]\n- if isinstance(eval_values[arg], pd.Series)\n- else eval_values[arg][indx]\n- )\n- else:\n- raise MlflowException(\n- f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n- )\n \n- return (\n- \"\"\n- if args_dict is None or len(args_dict) == 0\n- else (\n- \"Additional information used by the model:\\n\"\n- + \"\\n\".join(\n- [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n- )\n- )\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # Function to extract Score and Justification\n-def _extract_score_and_justification(text):\n- if text:\n- text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n- text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n- # Attempt to parse JSON\n- try:\n- data = json.loads(text)\n- score = int(data.get(\"score\"))\n- justification = data.get(\"justification\")\n- except json.JSONDecodeError:\n- # If parsing fails, use regex\n- if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n- match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n- ):\n- score = int(match.group(1))\n- justification = match.group(2)\n- else:\n- score = None\n- justification = f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- if not isinstance(score, (int, float)) or not isinstance(justification, str):\n- return None, f\"Failed to extract score and justification. Raw output: {text}\"\n-\n- return score, justification\n-\n- return None, None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n-def _score_model_on_payloads(\n- grading_payloads, model, parameters, headers, proxy_url, max_workers\n-) -> tuple[list[int], list[str]]:\n- scores = [None] * len(grading_payloads)\n- justifications = [None] * len(grading_payloads)\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- model,\n- parameters,\n- headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n-\n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- return scores, justifications\n-\n-\n-def _get_aggregate_results(scores, aggregations):\n- # loop over the aggregations and compute the aggregate results on the scores\n- def aggregate_function(aggregate_option, scores):\n- import numpy as np\n-\n- options = {\n- \"min\": np.min,\n- \"max\": np.max,\n- \"mean\": np.mean,\n- \"median\": np.median,\n- \"variance\": np.var,\n- \"p90\": lambda x: np.percentile(x, 90) if x else None,\n- }\n-\n- if aggregate_option not in options:\n- raise MlflowException(\n- message=f\"Invalid aggregate option {aggregate_option}.\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- return options[aggregate_option](scores)\n \n- scores_for_aggregation = [score for score in scores if score is not None]\n \n- return (\n- {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n- if aggregations is not None\n- else {}\n- )\n \n \n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric_from_prompt(\n- name: str,\n- judge_prompt: str | None = None,\n- model: str | None = _get_default_model(),\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n- a metric using only the supplied judge prompt without any pre-written system prompt.\n- This can be useful for use cases that are not covered by the full grading prompt in any\n- ``EvaluationModel`` version.\n \n- Args:\n- name: Name of the metric.\n- judge_prompt: The entire prompt to be used for the judge model.\n- The prompt will be minimally wrapped in formatting instructions to ensure\n- scores can be parsed. The prompt may use f-string formatting to include variables.\n- Corresponding variables must be passed as keyword arguments into the\n- resulting metric's eval function.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n \n- import pandas as pd\n- import mlflow\n- from mlflow.metrics.genai import make_genai_metric_from_prompt\n-\n- metric = make_genai_metric_from_prompt(\n- name=\"ease_of_understanding\",\n- judge_prompt=(\n- \"You must evaluate the output of a bot based on how easy it is to \"\n- \"understand its outputs.\"\n- \"Evaluate the bot's output from the perspective of a layperson.\"\n- \"The bot was provided with this input: {input} and this output: {output}.\"\n- ),\n- model=\"openai:/gpt-4\",\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n \n- data = pd.DataFrame(\n- {\n- \"input\": [\"Where is the capital of France.\"],\n- \"ground_truth\": [\"Paris\"],\n- \"output\": [\"The capital of France is Paris.\"],\n- }\n- )\n \n- mlflow.evaluate(\n- data=data,\n- targets=\"ground_truth\",\n- predictions=\"output\",\n- evaluators=\"default\",\n- extra_metrics=[metric],\n- )\n- \"\"\"\n- import numpy as np\n-\n- prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n- allowed_variables = prompt_template.variables\n- additional_variables = list(allowed_variables - {\"predictions\"})\n-\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"judge_prompt\": judge_prompt,\n- \"model\": model,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n-\n- def eval_fn(\n- *args,\n- **kwargs,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- Note that default evaluator only passes positional arguments.\n- \"\"\"\n- for i, arg in enumerate(args):\n- if i == 0:\n- kwargs[\"predictions\"] = arg\n- else:\n- kwargs[additional_variables[i - 1]] = arg\n- if missing_variables := allowed_variables - set(kwargs.keys()):\n- raise MlflowException(\n- message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n- kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n- grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n- arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n- scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n- )\n \n- aggregate_scores = _get_aggregate_results(scores, aggregations)\n-\n- return MetricValue(scores, justifications, aggregate_scores)\n-\n- # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n- eval_fn_parameters = [\n- Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n- *[\n- Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n- for var in additional_variables\n- ],\n- ]\n- eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n-\n- return make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n-\n-\n-@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n-def make_genai_metric(\n- name: str,\n- definition: str,\n- grading_prompt: str,\n- examples: list[EvaluationExample] | None = None,\n- version: str | None = _get_latest_metric_version(),\n- model: str | None = _get_default_model(),\n- grading_context_columns: str | list[str] | None = None,\n- include_input: bool = True,\n- parameters: dict[str, Any] | None = None,\n- aggregations: list[str] | None = None,\n- greater_is_better: bool = True,\n- max_workers: int = 10,\n- metric_metadata: dict[str, Any] | None = None,\n- extra_headers: dict[str, str] | None = None,\n- proxy_url: str | None = None,\n-) -> EvaluationMetric:\n- \"\"\"\n- Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n- prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n- Args:\n- name: Name of the metric.\n- definition: Definition of the metric.\n- grading_prompt: Grading criteria of the metric.\n- examples: (Optional) Examples of the metric.\n- version: (Optional) Version of the metric. Currently supported versions are: v1.\n- model: (Optional) Model uri of the judge model that will be used to compute the metric,\n- e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n- documentation for the supported model types and their URI format.\n- grading_context_columns: (Optional) The name of the grading context column, or a list of\n- grading context column names, required to compute the metric. The\n- ``grading_context_columns`` are used by the LLM as a judge as additional information to\n- compute the metric. The columns are extracted from the input dataset or output\n- predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n- :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n- include_input: (Optional) Whether to include the input\n- when computing the metric.\n- parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n- set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n- setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n- aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n- options are: min, max, mean, median, variance, p90.\n- greater_is_better: (Optional) Whether the metric is better when it is greater.\n- max_workers: (Optional) The maximum number of workers to use for judge scoring.\n- Defaults to 10 workers.\n- metric_metadata: (Optional) Dictionary of metadata to be attached to the\n- EvaluationMetric object. Useful for model evaluators that require additional\n- information to determine how to evaluate this metric.\n- extra_headers: (Optional) Additional headers to be passed to the judge model.\n- proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n- judge model is served via a proxy endpoint, not directly via LLM provider services.\n- If not specified, the default URL for the LLM provider will be used\n- (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n- Returns:\n- A metric object.\n \n- .. code-block:: python\n- :test:\n- :caption: Example for creating a genai metric\n-\n- from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n-\n- example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=(\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n- ),\n- score=4,\n- justification=(\n- \"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- ),\n- grading_context={\n- \"targets\": (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n- \"Databricks, a company that specializes in big data and machine learning \"\n- \"solutions. MLflow is designed to address the challenges that data \"\n- \"scientists and machine learning engineers face when developing, training, \"\n- \"and deploying machine learning models.\"\n- )\n- },\n- )\n- metric = make_genai_metric(\n- name=\"answer_correctness\",\n- definition=(\n- \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n- \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n- \"the degree of semantic similarity and factual correctness of the provided output \"\n- \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n- ),\n- grading_prompt=(\n- \"Answer correctness: Below are the details for different scores:\"\n- \"- Score 1: The output is completely incorrect. It is completely different from \"\n- \"or contradicts the provided targets.\"\n- \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n- \"includes partially correct information. However, the output still has significant \"\n- \"discrepancies with the provided targets or inaccuracies.\"\n- \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n- \"aligning with the provided targets. However, there are still omissions or minor \"\n- \"inaccuracies.\"\n- \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n- \"but there may be one or more minor omissions or inaccuracies.\"\n- \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n- \"semantic similarity to the targets.\"\n- ),\n- examples=[example],\n- version=\"v1\",\n- model=\"openai:/gpt-4\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- greater_is_better=True,\n- )\n- \"\"\"\n- # When users create a custom metric using this function,the metric configuration\n- # will be serialized and stored as an artifact. This enables us to later deserialize\n- # the configuration, allowing users to understand their LLM evaluation results more clearly.\n- genai_metric_args = {\n- \"name\": name,\n- \"definition\": definition,\n- \"grading_prompt\": grading_prompt,\n- \"examples\": examples,\n- \"version\": version,\n- \"model\": model,\n- \"grading_context_columns\": grading_context_columns,\n- \"include_input\": include_input,\n- \"parameters\": parameters,\n- \"aggregations\": aggregations,\n- \"greater_is_better\": greater_is_better,\n- \"max_workers\": max_workers,\n- \"metric_metadata\": metric_metadata,\n- # Record the mlflow version for serialization in case the function signature changes later\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n- grading_context_columns = grading_context_columns or []\n-\n- if not isinstance(grading_context_columns, list):\n- grading_context_columns = [grading_context_columns]\n-\n- def process_example(example):\n- if example.grading_context is None and len(grading_context_columns) == 0:\n- grading_context = {}\n- elif isinstance(example.grading_context, dict):\n- grading_context = example.grading_context\n- else:\n- # The grading context is string-like. Assume that it corresponds to the first\n- # grading context column and update the example accordingly\n- grading_context = {grading_context_columns[0]: example.grading_context}\n- example.grading_context = grading_context\n-\n- if set(grading_context.keys()) != set(grading_context_columns):\n- raise MlflowException.invalid_parameter_value(\n- f\"Example grading context does not contain required columns.\\n\"\n- f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n- f\" Required grading context columns: {grading_context_columns}\\n\"\n- )\n \n- if not include_input:\n- return EvaluationExample(\n- output=example.output,\n- score=example.score,\n- justification=example.justification,\n- grading_context=example.grading_context,\n- )\n \n- return example\n \n- if examples is not None:\n- examples = [process_example(example) for example in examples]\n \n- class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n- try:\n- evaluation_model_class_module = _get_class_from_string(class_name)\n- except ModuleNotFoundError:\n- raise MlflowException(\n- f\"Failed to find evaluation model for version {version}.\"\n- f\" Please check the correctness of the version\",\n- error_code=INVALID_PARAMETER_VALUE,\n- ) from None\n- except Exception as e:\n- raise MlflowException(\n- f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n- error_code=INTERNAL_ERROR,\n- ) from None\n-\n- evaluation_context = evaluation_model_class_module(\n- name,\n- definition,\n- grading_prompt,\n- examples,\n- model,\n- *(parameters,) if parameters is not None else (),\n- include_input=include_input,\n- ).to_dict()\n-\n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- \"\"\"\n- This is the function that is called when the metric is evaluated.\n- \"\"\"\n- eval_values = dict(zip(grading_context_columns, args))\n-\n- outputs = predictions.to_list()\n- inputs = inputs.to_list()\n- eval_model = evaluation_context[\"model\"]\n- eval_parameters = evaluation_context[\"parameters\"]\n-\n- # TODO: Save the metric definition in a yaml file for model monitoring\n-\n- if not isinstance(eval_model, str):\n- raise MlflowException(\n- message=\"The model argument must be a string URI referring to an openai model \"\n- \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n- f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n- error_code=INVALID_PARAMETER_VALUE,\n- )\n \n- # generate grading payloads\n- grading_payloads = []\n- for indx, (input, output) in enumerate(zip(inputs, outputs)):\n- try:\n- arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n- except Exception as e:\n- raise MlflowException(\n- f\"Values for grading_context_columns are malformed and cannot be \"\n- f\"formatted into a prompt for metric '{name}'.\\n\"\n- f\"Required columns: {grading_context_columns}\\n\"\n- f\"Values: {eval_values}\\n\"\n- f\"Error: {e!r}\\n\"\n- f\"Please check the following: \\n\"\n- \"- predictions and targets (if required) are provided correctly\\n\"\n- \"- grading_context_columns are mapped correctly using the evaluator_config \"\n- \"parameter\\n\"\n- \"- input and output data are formatted correctly.\"\n- )\n- grading_payloads.append(\n- evaluation_context[\"eval_prompt\"].format(\n- input=(input if include_input else None),\n- output=output,\n- grading_context_columns=arg_string,\n- )\n- )\n \n- scores = [None] * len(inputs)\n- justifications = [None] * len(inputs)\n-\n- with ThreadPoolExecutor(\n- max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n- ) as executor:\n- futures = {\n- executor.submit(\n- _score_model_on_one_payload,\n- payload,\n- eval_model,\n- eval_parameters,\n- extra_headers,\n- proxy_url,\n- ): indx\n- for indx, payload in enumerate(grading_payloads)\n- }\n \n- as_comp = as_completed(futures)\n- try:\n- from tqdm.auto import tqdm\n-\n- as_comp = tqdm(as_comp, total=len(futures))\n- except ImportError:\n- pass\n-\n- for future in as_comp:\n- indx = futures[future]\n- score, justification = future.result()\n- scores[indx] = score\n- justifications[indx] = justification\n-\n- aggregate_results = _get_aggregate_results(scores, aggregations)\n- return MetricValue(scores, justifications, aggregate_results)\n-\n- signature_parameters = [\n- Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n- Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n- ]\n-\n- # Add grading_context_columns to signature list\n- for var in grading_context_columns:\n- signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n-\n- # Note: this doesn't change how python allows calling the function\n- # extra params in grading_context_columns can only be passed as positional args\n- eval_fn.__signature__ = Signature(signature_parameters)\n-\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- version=version,\n- metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=True,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n--- a/mlflow/metrics/genai/metric_definitions.py\n+++ b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n-from mlflow.metrics.genai.genai_metric import make_genai_metric\n+\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n--- a/mlflow/metrics/genai/prompt_template.py\n+++ b/mlflow/metrics/genai/prompt_template.py\n@@ -29,40 +29,36 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n- @property\n- def variables(self):\n- return {\n- fname\n- for template_str in self.template_strs\n- for _, fname, _, _ in string.Formatter().parse(template_str)\n- if fname\n- }\n-\n- def format(self, **kwargs: Any) -> str:\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- formatted_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- if all(item in safe_kwargs.keys() for item in extracted_variables):\n- formatted_strs.append(template_str.format(**safe_kwargs))\n-\n- return \"\".join(formatted_strs)\n-\n- def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n- safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n- new_template_strs = []\n- for template_str in self.template_strs:\n- extracted_variables = [\n- fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n- ]\n- safe_available_kwargs = {\n- k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n- }\n- new_template_strs.append(template_str.format_map(safe_available_kwargs))\n-\n- return PromptTemplate(template_str=new_template_strs)\n-\n- def __str__(self) -> str:\n- return \"\".join(self.template_strs)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n--- a/mlflow/metrics/genai/prompts/v1.py\n+++ b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n- def to_dict(self):\n- examples_str = (\n- \"\"\n- if self.examples is None or len(self.examples) == 0\n- else f\"Examples:\\n{self._format_examples()}\"\n- )\n \n- template = _build_grading_prompt_template(include_input=self.include_input)\n \n- return {\n- \"model\": self.model,\n- \"eval_prompt\": template.partial_fill(\n- name=self.name,\n- definition=self.definition,\n- grading_prompt=self.grading_prompt,\n- examples=examples_str,\n- ),\n- \"parameters\": self.parameters,\n- }\n \n- def _format_examples(self):\n- return \"\\n\".join(map(str, self.examples))\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n- default_examples = [example_score_2, example_score_4]\n+ default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n--- a/mlflow/models/__init__.py\n+++ b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n-from mlflow.models.evaluation import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- MetricThreshold,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n--- a/mlflow/models/evaluation/__init__.py\n+++ b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n-from mlflow.models.evaluation.base import (\n- EvaluationArtifact,\n- EvaluationMetric,\n- EvaluationResult,\n- ModelEvaluator,\n- evaluate,\n- list_evaluators,\n- make_metric,\n-)\n+from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n+\n+\n+\n+\n+\n+\n+\n+\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n--- a/mlflow/models/evaluation/base.py\n+++ b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n-from mlflow.utils.class_utils import _get_class_from_string\n+\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n- def __str__(self):\n- parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n- if self.long_name:\n- parts.append(f\"long_name={self.long_name}\")\n- if self.version:\n- parts.append(f\"version={self.version}\")\n- if self.metric_details:\n- parts.append(f\"metric_details={self.metric_details}\")\n- if self.metric_metadata:\n- parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n- return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n-def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n- \"\"\"\n- Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n- on the given input data. The generated class is callable with a __call__ method that\n- takes the arguments specified in the signature of the eval_fn function.\n \n- Args:\n- eval_fn: the evaluation function of the EvaluationMetric.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- Returns:\n- A dynamically generated callable CallableEvaluationMetric class.\n- \"\"\"\n- from mlflow.metrics.base import MetricValue\n \n- if require_strict_signature:\n- allowed_kwargs_names = [\n- param_name\n- for param_name in inspect.signature(eval_fn).parameters.keys()\n- if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n- ]\n \n- def genai_call_method(\n- self,\n- *,\n- predictions: pd.Series | str | list[str],\n- inputs: pd.Series | str | list[str],\n- metrics: dict[str, MetricValue] | None = None,\n- **kwargs,\n- ) -> MetricValue:\n- if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n- raise MlflowException.invalid_parameter_value(\n- f\"Missing required arguments: {missed_kwargs}\",\n- )\n- if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n- raise MlflowException.invalid_parameter_value(\n- f\"Unexpected arguments: {extra_kwargs}\",\n- )\n- return self.eval_fn(\n- _convert_val_to_pd_Series(predictions, \"predictions\"),\n- metrics or {},\n- _convert_val_to_pd_Series(inputs, \"inputs\"),\n- # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n- # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n- # should always be pandas Series\n- *[\n- _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n- for arg_name in allowed_kwargs_names\n- ],\n- )\n \n- genai_call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- Parameter(\n- \"predictions\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"inputs\",\n- Parameter.KEYWORD_ONLY,\n- annotation=pd.Series | str | list[str],\n- ),\n- Parameter(\n- \"metrics\",\n- Parameter.KEYWORD_ONLY,\n- annotation=dict[str, MetricValue] | None,\n- default=None,\n- ),\n- *[\n- Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n- for name in allowed_kwargs_names\n- ],\n- ]\n- )\n- genai_call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- predictions: predictions made by the model.\n- inputs: inputs used to make the predictions.\n- metrics: metrics calculated by the default evaluator.\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {allowed_kwargs_names}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = genai_call_method\n \n- else:\n \n- def _call_method(\n- self,\n- **kwargs,\n- ) -> MetricValue:\n- return self.eval_fn(**kwargs)\n-\n- allowed_kwargs_params = inspect.signature(eval_fn).parameters\n- _call_method.__signature__ = Signature(\n- parameters=[\n- Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n- *[\n- Parameter(\n- name,\n- Parameter.KEYWORD_ONLY,\n- annotation=allowed_kwargs_params[name].annotation,\n- )\n- for name in allowed_kwargs_params.keys()\n- ],\n- ]\n- )\n- _call_method.__doc__ = f\"\"\"\n- Evaluate the metric on the given inputs and predictions.\n- Note: only keyword arguments are supported.\n-\n- Args:\n- kwargs: additional arguments used to compute the metric.\n- Required arguments: {list(allowed_kwargs_params.keys())}\n-\n- Returns:\n- evaluation result as MetricValue object.\n- \"\"\"\n- call_method = _call_method\n-\n- return type(\n- \"CallableEvaluationMetric\",\n- (EvaluationMetric,),\n- {\"__call__\": call_method},\n- )\n \n \n-def _convert_val_to_pd_Series(val, name):\n- if val is not None and not isinstance(val, pd.Series):\n- if isinstance(val, str):\n- return pd.Series([val])\n- elif isinstance(val, list):\n- return pd.Series(val)\n- else:\n- raise TypeError(\n- f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n- )\n- return val\n \n \n-def make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- return _make_metric(\n- eval_fn=eval_fn,\n- greater_is_better=greater_is_better,\n- name=name,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- require_strict_signature=False,\n- )\n \n \n-def _make_metric(\n- *,\n- eval_fn,\n- greater_is_better,\n- name=None,\n- long_name=None,\n- version=None,\n- metric_details=None,\n- metric_metadata=None,\n- genai_metric_args=None,\n- require_strict_signature=False,\n-):\n- '''\n- A factory function to create an :py:class:`EvaluationMetric` object.\n \n- Args:\n- eval_fn: A function that computes the metric with the following signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: pandas.Series,\n- targets: pandas.Series,\n- metrics: Dict[str, MetricValue],\n- **kwargs,\n- ) -> Union[float, MetricValue]:\n- \"\"\"\n- Args:\n- predictions: A pandas Series containing the predictions made by the model.\n- targets: (Optional) A pandas Series containing the corresponding labels\n- for the predictions made on that input.\n- metrics: (Optional) A dictionary containing the metrics calculated by the\n- default evaluator. The keys are the names of the metrics and the values\n- are the metric values. To access the MetricValue for the metrics\n- calculated by the system, make sure to specify the type hint for this\n- parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n- behavior section for what metrics will be returned based on the type of\n- model (i.e. classifier or regressor). kwargs: Includes a list of args\n- that are used to compute the metric. These args could information coming\n- from input data, model outputs or parameters specified in the\n- `evaluator_config` argument of the `mlflow.evaluate` API.\n- kwargs: Includes a list of args that are used to compute the metric. These\n- args could be information coming from input data, model outputs,\n- other metrics, or parameters specified in the `evaluator_config`\n- argument of the `mlflow.evaluate` API.\n \n- Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n- results.\n- \"\"\"\n- ...\n \n- greater_is_better: Whether a greater value of the metric is better.\n- name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n- function or the ``eval_fn.__name__`` attribute is not available.\n- long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n- for ``\"mse\"``.\n- version: (Optional) The metric version. For example ``v1``.\n- metric_details: (Optional) A description of the metric and how it is calculated.\n- metric_metadata: (Optional) A dictionary containing metadata for the metric.\n- genai_metric_args: (Optional) A dictionary containing arguments specified by users\n- when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n- are persisted so that we can deserialize the same metric object later.\n- require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n- If True, then the eval_fn must follow below signature:\n \n- .. code-block:: python\n \n- def eval_fn(\n- predictions: \"pd.Series\",\n- metrics: Dict[str, MetricValue],\n- inputs: \"pd.Series\",\n- *args,\n- ) -> MetricValue:\n- pass\n \n- When generating a metric from `make_genai_metric`, this should be set to True.\n- Default to False.\n \n- .. seealso::\n \n- - :py:class:`mlflow.models.EvaluationMetric`\n- - :py:func:`mlflow.evaluate`\n- '''\n- if name is None:\n- if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` is a lambda function.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- if not hasattr(eval_fn, \"__name__\"):\n- raise MlflowException(\n- \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n- INVALID_PARAMETER_VALUE,\n- )\n- name = eval_fn.__name__\n \n- if \"/\" in name:\n- raise MlflowException(\n- f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n- INVALID_PARAMETER_VALUE,\n- )\n \n- if not name.isidentifier():\n- _logger.warning(\n- f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if keyword.iskeyword(name):\n- _logger.warning(\n- f\"The metric name '{name}' is a reserved Python keyword, which will \"\n- \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n- \"to enable creation of derived metrics that use the given metric.\"\n- )\n \n- if name in [\"predictions\", \"targets\", \"metrics\"]:\n- _logger.warning(\n- f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n- \"will prevent its use as a base metric for derived metrics. Please use a different \"\n- \"name to enable creation of derived metrics that use the given metric.\"\n- )\n \n- return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n- eval_fn=eval_fn,\n- name=name,\n- greater_is_better=greater_is_better,\n- long_name=long_name,\n- version=version,\n- metric_details=metric_details,\n- metric_metadata=metric_metadata,\n- genai_metric_args=genai_metric_args,\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n- def __init__(self, metrics, artifacts, run_id=None):\n- self._metrics = metrics\n- self._artifacts = artifacts\n- self._run_id = (\n- run_id\n- if run_id is not None\n- else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n- return evaluate_result\n+ return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n--- a/mlflow/utils/class_utils.py\n+++ b/mlflow/utils/class_utils.py\n@@ -1,6 +1,4 @@\n import importlib\n \n \n-def _get_class_from_string(fully_qualified_class_name):\n- module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n- return getattr(importlib.import_module(module), class_name)\n+\ndiff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\nindex 5770256f1..e98198036 100644\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ b/tests/metrics/genai/test_genai_metrics.py\n@@ -8,12 +8,12 @@ import pytest\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n+\n+\n+\n+\n+\n+\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py\nindex de8928c86..56c409983 100644\n+++ a/mlflow/metrics/__init__.py\n--- b/mlflow/metrics/__init__.py\n@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (\n _rougeLsum_eval_fn,\n _toxicity_eval_fn,\n )\n+from mlflow.models import (\n+ EvaluationMetric,\n+ make_metric,\n+)\n-from mlflow.models import EvaluationMetric\n-\n-\n-\n from mlflow.utils.annotations import deprecated\n \n \ndiff --git a/mlflow/metrics/base.py b/mlflow/metrics/base.py\nindex c998e559c..59eaf1035 100644\n+++ a/mlflow/metrics/base.py\n--- b/mlflow/metrics/base.py\n@@ -29,8 +29,10 @@ class MetricValue:\n justifications: list[str] | None = None\n aggregate_results: dict[str, float] | None = None\n \n+ def __post_init__(self):\n+ if (\n+ self.aggregate_results is None\n+ and isinstance(self.scores, (list, tuple))\n+ and all(_is_numeric(score) for score in self.scores)\n+ ):\n+ self.aggregate_results = standard_aggregations(self.scores)\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/__init__.py b/mlflow/metrics/genai/__init__.py\nindex 32604503e..bfd635f82 100644\n+++ a/mlflow/metrics/genai/__init__.py\n--- b/mlflow/metrics/genai/__init__.py\n@@ -1,9 +1,9 @@\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import (\n+ make_genai_metric,\n+ make_genai_metric_from_prompt,\n+ retrieve_custom_metrics,\n+)\n-from mlflow.metrics.genai.genai_metric import retrieve_custom_metrics\n-\n-\n-\n-\n from mlflow.metrics.genai.metric_definitions import (\n answer_correctness,\n answer_relevance,\ndiff --git a/mlflow/metrics/genai/base.py b/mlflow/metrics/genai/base.py\nindex 05e67f638..f87ae3b40 100644\n+++ a/mlflow/metrics/genai/base.py\n--- b/mlflow/metrics/genai/base.py\n@@ -63,35 +63,38 @@ class EvaluationExample:\n input: str | None = None\n grading_context: dict[str, str] | str | None = None\n \n+ def _format_grading_context(self):\n+ if isinstance(self.grading_context, dict):\n+ return \"\\n\".join(\n+ [f\"key: {key}\\nvalue:\\n{value}\" for key, value in self.grading_context.items()]\n+ )\n+ else:\n+ return self.grading_context\n+\n+ def __str__(self) -> str:\n+ return PromptTemplate(\n+ [\n+ \"\"\"\n+Example Input:\n+{input}\n+\"\"\",\n+ \"\"\"\n+Example Output:\n+{output}\n+\"\"\",\n+ \"\"\"\n+Additional information used by the model:\n+{grading_context}\n+\"\"\",\n+ \"\"\"\n+Example score: {score}\n+Example justification: {justification}\n+ \"\"\",\n+ ]\n+ ).format(\n+ input=self.input,\n+ output=self.output,\n+ grading_context=self._format_grading_context(),\n+ score=self.score,\n+ justification=self.justification,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..c5336ba2a 100644\n+++ a/mlflow/metrics/genai/genai_metric.py\n--- b/mlflow/metrics/genai/genai_metric.py\n@@ -20,8 +20,8 @@ from mlflow.metrics.genai.utils import (\n _get_default_model,\n _get_latest_metric_version,\n )\n+from mlflow.models import EvaluationMetric, make_metric\n+from mlflow.models.evaluation.base import _make_metric\n-from mlflow.models import EvaluationMetric\n-\n from mlflow.protos.databricks_pb2 import (\n BAD_REQUEST,\n INTERNAL_ERROR,\n@@ -30,7 +30,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.version import VERSION\n \n _logger = logging.getLogger(__name__)\n@@ -45,61 +45,61 @@ justification: Your reasoning for giving this score\n Do not add additional new lines. Do not add any other fields.\"\"\"\n \n \n+def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str:\n+ import pandas as pd\n \n+ args_dict = {}\n+ for arg in grading_context_columns:\n+ if arg in eval_values:\n+ args_dict[arg] = (\n+ eval_values[arg].iloc[indx]\n+ if isinstance(eval_values[arg], pd.Series)\n+ else eval_values[arg][indx]\n+ )\n+ else:\n+ raise MlflowException(\n+ f\"{arg} does not exist in the eval function {list(eval_values.keys())}.\"\n+ )\n \n+ return (\n+ \"\"\n+ if args_dict is None or len(args_dict) == 0\n+ else (\n+ \"Additional information used by the model:\\n\"\n+ + \"\\n\".join(\n+ [f\"key: {arg}\\nvalue:\\n{arg_value}\" for arg, arg_value in args_dict.items()]\n+ )\n+ )\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # Function to extract Score and Justification\n+def _extract_score_and_justification(text):\n+ if text:\n+ text = re.sub(r\"score\", \"score\", text, flags=re.IGNORECASE)\n+ text = re.sub(r\"justification\", \"justification\", text, flags=re.IGNORECASE)\n+ # Attempt to parse JSON\n+ try:\n+ data = json.loads(text)\n+ score = int(data.get(\"score\"))\n+ justification = data.get(\"justification\")\n+ except json.JSONDecodeError:\n+ # If parsing fails, use regex\n+ if (match := re.search(r\"score: (\\d+),?\\s*justification: (.+)\", text)) or (\n+ match := re.search(r\"\\s*score:\\s*(\\d+)\\s*justification:\\s*(.+)\", text, re.DOTALL)\n+ ):\n+ score = int(match.group(1))\n+ justification = match.group(2)\n+ else:\n+ score = None\n+ justification = f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ if not isinstance(score, (int, float)) or not isinstance(justification, str):\n+ return None, f\"Failed to extract score and justification. Raw output: {text}\"\n+\n+ return score, justification\n+\n+ return None, None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _score_model_on_one_payload(\n@@ -131,550 +131,550 @@ def _score_model_on_one_payload(\n return None, f\"Failed to score model on payload. Error: {e!s}\"\n \n \n+def _score_model_on_payloads(\n+ grading_payloads, model, parameters, headers, proxy_url, max_workers\n+) -> tuple[list[int], list[str]]:\n+ scores = [None] * len(grading_payloads)\n+ justifications = [None] * len(grading_payloads)\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiScoring\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ model,\n+ parameters,\n+ headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n+\n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ return scores, justifications\n+\n+\n+def _get_aggregate_results(scores, aggregations):\n+ # loop over the aggregations and compute the aggregate results on the scores\n+ def aggregate_function(aggregate_option, scores):\n+ import numpy as np\n+\n+ options = {\n+ \"min\": np.min,\n+ \"max\": np.max,\n+ \"mean\": np.mean,\n+ \"median\": np.median,\n+ \"variance\": np.var,\n+ \"p90\": lambda x: np.percentile(x, 90) if x else None,\n+ }\n+\n+ if aggregate_option not in options:\n+ raise MlflowException(\n+ message=f\"Invalid aggregate option {aggregate_option}.\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ return options[aggregate_option](scores)\n \n+ scores_for_aggregation = [score for score in scores if score is not None]\n \n+ return (\n+ {option: aggregate_function(option, scores_for_aggregation) for option in aggregations}\n+ if aggregations is not None\n+ else {}\n+ )\n \n \n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric_from_prompt(\n+ name: str,\n+ judge_prompt: str | None = None,\n+ model: str | None = _get_default_model(),\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. This produces\n+ a metric using only the supplied judge prompt without any pre-written system prompt.\n+ This can be useful for use cases that are not covered by the full grading prompt in any\n+ ``EvaluationModel`` version.\n \n+ Args:\n+ name: Name of the metric.\n+ judge_prompt: The entire prompt to be used for the judge model.\n+ The prompt will be minimally wrapped in formatting instructions to ensure\n+ scores can be parsed. The prompt may use f-string formatting to include variables.\n+ Corresponding variables must be passed as keyword arguments into the\n+ resulting metric's eval function.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n \n+ import pandas as pd\n+ import mlflow\n+ from mlflow.metrics.genai import make_genai_metric_from_prompt\n+\n+ metric = make_genai_metric_from_prompt(\n+ name=\"ease_of_understanding\",\n+ judge_prompt=(\n+ \"You must evaluate the output of a bot based on how easy it is to \"\n+ \"understand its outputs.\"\n+ \"Evaluate the bot's output from the perspective of a layperson.\"\n+ \"The bot was provided with this input: {input} and this output: {output}.\"\n+ ),\n+ model=\"openai:/gpt-4\",\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n \n+ data = pd.DataFrame(\n+ {\n+ \"input\": [\"Where is the capital of France.\"],\n+ \"ground_truth\": [\"Paris\"],\n+ \"output\": [\"The capital of France is Paris.\"],\n+ }\n+ )\n \n+ mlflow.evaluate(\n+ data=data,\n+ targets=\"ground_truth\",\n+ predictions=\"output\",\n+ evaluators=\"default\",\n+ extra_metrics=[metric],\n+ )\n+ \"\"\"\n+ import numpy as np\n+\n+ prompt_template = PromptTemplate([judge_prompt, _PROMPT_FORMATTING_WRAPPER])\n+ allowed_variables = prompt_template.variables\n+ additional_variables = list(allowed_variables - {\"predictions\"})\n+\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"judge_prompt\": judge_prompt,\n+ \"model\": model,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric_from_prompt.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+\n+ def eval_fn(\n+ *args,\n+ **kwargs,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ Note that default evaluator only passes positional arguments.\n+ \"\"\"\n+ for i, arg in enumerate(args):\n+ if i == 0:\n+ kwargs[\"predictions\"] = arg\n+ else:\n+ kwargs[additional_variables[i - 1]] = arg\n+ if missing_variables := allowed_variables - set(kwargs.keys()):\n+ raise MlflowException(\n+ message=f\"Missing variable inputs to eval_fn: {missing_variables}\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n+ kwargs = {k: [v] if np.isscalar(v) else v for k, v in kwargs.items()}\n+ grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n+ arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n+ scores, justifications = _score_model_on_payloads(\n+ arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ )\n \n+ aggregate_scores = _get_aggregate_results(scores, aggregations)\n+\n+ return MetricValue(scores, justifications, aggregate_scores)\n+\n+ # Add `predictions` to the parameters to be compatible with `eval_fn`` interface\n+ eval_fn_parameters = [\n+ Parameter(name=\"predictions\", kind=Parameter.POSITIONAL_ONLY),\n+ *[\n+ Parameter(name=var, kind=Parameter.POSITIONAL_ONLY, default=None)\n+ for var in additional_variables\n+ ],\n+ ]\n+ eval_fn.__signature__ = Signature(parameters=eval_fn_parameters)\n+\n+ return make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n+\n+\n+@deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\n+def make_genai_metric(\n+ name: str,\n+ definition: str,\n+ grading_prompt: str,\n+ examples: list[EvaluationExample] | None = None,\n+ version: str | None = _get_latest_metric_version(),\n+ model: str | None = _get_default_model(),\n+ grading_context_columns: str | list[str] | None = None,\n+ include_input: bool = True,\n+ parameters: dict[str, Any] | None = None,\n+ aggregations: list[str] | None = None,\n+ greater_is_better: bool = True,\n+ max_workers: int = 10,\n+ metric_metadata: dict[str, Any] | None = None,\n+ extra_headers: dict[str, str] | None = None,\n+ proxy_url: str | None = None,\n+) -> EvaluationMetric:\n+ \"\"\"\n+ Create a genai metric used to evaluate LLM using LLM as a judge in MLflow. The full grading\n+ prompt is stored in the metric_details field of the ``EvaluationMetric`` object.\n \n+ Args:\n+ name: Name of the metric.\n+ definition: Definition of the metric.\n+ grading_prompt: Grading criteria of the metric.\n+ examples: (Optional) Examples of the metric.\n+ version: (Optional) Version of the metric. Currently supported versions are: v1.\n+ model: (Optional) Model uri of the judge model that will be used to compute the metric,\n+ e.g., ``openai:/gpt-4``. Refer to the `LLM-as-a-Judge Metrics <https://mlflow.org/docs/latest/llms/llm-evaluate/index.html#selecting-the-llm-as-judge-model>`_\n+ documentation for the supported model types and their URI format.\n+ grading_context_columns: (Optional) The name of the grading context column, or a list of\n+ grading context column names, required to compute the metric. The\n+ ``grading_context_columns`` are used by the LLM as a judge as additional information to\n+ compute the metric. The columns are extracted from the input dataset or output\n+ predictions based on ``col_mapping`` in the ``evaluator_config`` passed to\n+ :py:func:`mlflow.evaluate()`. They can also be the name of other evaluated metrics.\n+ include_input: (Optional) Whether to include the input\n+ when computing the metric.\n+ parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we\n+ set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend\n+ setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.\n+ aggregations: (Optional) The list of options to aggregate the scores. Currently supported\n+ options are: min, max, mean, median, variance, p90.\n+ greater_is_better: (Optional) Whether the metric is better when it is greater.\n+ max_workers: (Optional) The maximum number of workers to use for judge scoring.\n+ Defaults to 10 workers.\n+ metric_metadata: (Optional) Dictionary of metadata to be attached to the\n+ EvaluationMetric object. Useful for model evaluators that require additional\n+ information to determine how to evaluate this metric.\n+ extra_headers: (Optional) Additional headers to be passed to the judge model.\n+ proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the\n+ judge model is served via a proxy endpoint, not directly via LLM provider services.\n+ If not specified, the default URL for the LLM provider will be used\n+ (e.g., https://api.openai.com/v1/chat/completions for OpenAI chat models).\n \n \n+ Returns:\n+ A metric object.\n \n+ .. code-block:: python\n+ :test:\n+ :caption: Example for creating a genai metric\n+\n+ from mlflow.metrics.genai import EvaluationExample, make_genai_metric\n+\n+ example = EvaluationExample(\n+ input=\"What is MLflow?\",\n+ output=(\n+ \"MLflow is an open-source platform for managing machine \"\n+ \"learning workflows, including experiment tracking, model packaging, \"\n+ \"versioning, and deployment, simplifying the ML lifecycle.\"\n+ ),\n+ score=4,\n+ justification=(\n+ \"The definition effectively explains what MLflow is \"\n+ \"its purpose, and its developer. It could be more concise for a 5-score.\",\n+ ),\n+ grading_context={\n+ \"targets\": (\n+ \"MLflow is an open-source platform for managing \"\n+ \"the end-to-end machine learning (ML) lifecycle. It was developed by \"\n+ \"Databricks, a company that specializes in big data and machine learning \"\n+ \"solutions. MLflow is designed to address the challenges that data \"\n+ \"scientists and machine learning engineers face when developing, training, \"\n+ \"and deploying machine learning models.\"\n+ )\n+ },\n+ )\n+ metric = make_genai_metric(\n+ name=\"answer_correctness\",\n+ definition=(\n+ \"Answer correctness is evaluated on the accuracy of the provided output based on \"\n+ \"the provided targets, which is the ground truth. Scores can be assigned based on \"\n+ \"the degree of semantic similarity and factual correctness of the provided output \"\n+ \"to the provided targets, where a higher score indicates higher degree of accuracy.\"\n+ ),\n+ grading_prompt=(\n+ \"Answer correctness: Below are the details for different scores:\"\n+ \"- Score 1: The output is completely incorrect. It is completely different from \"\n+ \"or contradicts the provided targets.\"\n+ \"- Score 2: The output demonstrates some degree of semantic similarity and \"\n+ \"includes partially correct information. However, the output still has significant \"\n+ \"discrepancies with the provided targets or inaccuracies.\"\n+ \"- Score 3: The output addresses a couple of aspects of the input accurately, \"\n+ \"aligning with the provided targets. However, there are still omissions or minor \"\n+ \"inaccuracies.\"\n+ \"- Score 4: The output is mostly correct. It provides mostly accurate information, \"\n+ \"but there may be one or more minor omissions or inaccuracies.\"\n+ \"- Score 5: The output is correct. It demonstrates a high degree of accuracy and \"\n+ \"semantic similarity to the targets.\"\n+ ),\n+ examples=[example],\n+ version=\"v1\",\n+ model=\"openai:/gpt-4\",\n+ grading_context_columns=[\"targets\"],\n+ parameters={\"temperature\": 0.0},\n+ aggregations=[\"mean\", \"variance\", \"p90\"],\n+ greater_is_better=True,\n+ )\n+ \"\"\"\n+ # When users create a custom metric using this function,the metric configuration\n+ # will be serialized and stored as an artifact. This enables us to later deserialize\n+ # the configuration, allowing users to understand their LLM evaluation results more clearly.\n+ genai_metric_args = {\n+ \"name\": name,\n+ \"definition\": definition,\n+ \"grading_prompt\": grading_prompt,\n+ \"examples\": examples,\n+ \"version\": version,\n+ \"model\": model,\n+ \"grading_context_columns\": grading_context_columns,\n+ \"include_input\": include_input,\n+ \"parameters\": parameters,\n+ \"aggregations\": aggregations,\n+ \"greater_is_better\": greater_is_better,\n+ \"max_workers\": max_workers,\n+ \"metric_metadata\": metric_metadata,\n+ # Record the mlflow version for serialization in case the function signature changes later\n+ \"mlflow_version\": VERSION,\n+ \"fn_name\": make_genai_metric.__name__,\n+ }\n+\n+ aggregations = aggregations or [\"mean\", \"variance\", \"p90\"]\n+ grading_context_columns = grading_context_columns or []\n+\n+ if not isinstance(grading_context_columns, list):\n+ grading_context_columns = [grading_context_columns]\n+\n+ def process_example(example):\n+ if example.grading_context is None and len(grading_context_columns) == 0:\n+ grading_context = {}\n+ elif isinstance(example.grading_context, dict):\n+ grading_context = example.grading_context\n+ else:\n+ # The grading context is string-like. Assume that it corresponds to the first\n+ # grading context column and update the example accordingly\n+ grading_context = {grading_context_columns[0]: example.grading_context}\n+ example.grading_context = grading_context\n+\n+ if set(grading_context.keys()) != set(grading_context_columns):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Example grading context does not contain required columns.\\n\"\n+ f\" Example grading context columns: {list(grading_context.keys())}\\n\"\n+ f\" Required grading context columns: {grading_context_columns}\\n\"\n+ )\n \n+ if not include_input:\n+ return EvaluationExample(\n+ output=example.output,\n+ score=example.score,\n+ justification=example.justification,\n+ grading_context=example.grading_context,\n+ )\n \n+ return example\n \n+ if examples is not None:\n+ examples = [process_example(example) for example in examples]\n \n+ class_name = f\"mlflow.metrics.genai.prompts.{version}.EvaluationModel\"\n+ try:\n+ evaluation_model_class_module = _get_class_from_string(class_name)\n+ except ModuleNotFoundError:\n+ raise MlflowException(\n+ f\"Failed to find evaluation model for version {version}.\"\n+ f\" Please check the correctness of the version\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ ) from None\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Failed to construct evaluation model {version}. Error: {e!r}\",\n+ error_code=INTERNAL_ERROR,\n+ ) from None\n+\n+ evaluation_context = evaluation_model_class_module(\n+ name,\n+ definition,\n+ grading_prompt,\n+ examples,\n+ model,\n+ *(parameters,) if parameters is not None else (),\n+ include_input=include_input,\n+ ).to_dict()\n+\n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ \"\"\"\n+ This is the function that is called when the metric is evaluated.\n+ \"\"\"\n+ eval_values = dict(zip(grading_context_columns, args))\n+\n+ outputs = predictions.to_list()\n+ inputs = inputs.to_list()\n+ eval_model = evaluation_context[\"model\"]\n+ eval_parameters = evaluation_context[\"parameters\"]\n+\n+ # TODO: Save the metric definition in a yaml file for model monitoring\n+\n+ if not isinstance(eval_model, str):\n+ raise MlflowException(\n+ message=\"The model argument must be a string URI referring to an openai model \"\n+ \"(openai:/gpt-4o-mini) or an MLflow Deployments endpoint \"\n+ f\"(endpoints:/my-endpoint), passed {eval_model} instead\",\n+ error_code=INVALID_PARAMETER_VALUE,\n+ )\n \n+ # generate grading payloads\n+ grading_payloads = []\n+ for indx, (input, output) in enumerate(zip(inputs, outputs)):\n+ try:\n+ arg_string = _format_args_string(grading_context_columns, eval_values, indx)\n+ except Exception as e:\n+ raise MlflowException(\n+ f\"Values for grading_context_columns are malformed and cannot be \"\n+ f\"formatted into a prompt for metric '{name}'.\\n\"\n+ f\"Required columns: {grading_context_columns}\\n\"\n+ f\"Values: {eval_values}\\n\"\n+ f\"Error: {e!r}\\n\"\n+ f\"Please check the following: \\n\"\n+ \"- predictions and targets (if required) are provided correctly\\n\"\n+ \"- grading_context_columns are mapped correctly using the evaluator_config \"\n+ \"parameter\\n\"\n+ \"- input and output data are formatted correctly.\"\n+ )\n+ grading_payloads.append(\n+ evaluation_context[\"eval_prompt\"].format(\n+ input=(input if include_input else None),\n+ output=output,\n+ grading_context_columns=arg_string,\n+ )\n+ )\n \n+ scores = [None] * len(inputs)\n+ justifications = [None] * len(inputs)\n+\n+ with ThreadPoolExecutor(\n+ max_workers=max_workers, thread_name_prefix=\"MlflowGenAiEvaluation\"\n+ ) as executor:\n+ futures = {\n+ executor.submit(\n+ _score_model_on_one_payload,\n+ payload,\n+ eval_model,\n+ eval_parameters,\n+ extra_headers,\n+ proxy_url,\n+ ): indx\n+ for indx, payload in enumerate(grading_payloads)\n+ }\n \n+ as_comp = as_completed(futures)\n+ try:\n+ from tqdm.auto import tqdm\n+\n+ as_comp = tqdm(as_comp, total=len(futures))\n+ except ImportError:\n+ pass\n+\n+ for future in as_comp:\n+ indx = futures[future]\n+ score, justification = future.result()\n+ scores[indx] = score\n+ justifications[indx] = justification\n+\n+ aggregate_results = _get_aggregate_results(scores, aggregations)\n+ return MetricValue(scores, justifications, aggregate_results)\n+\n+ signature_parameters = [\n+ Parameter(\"predictions\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ Parameter(\"metrics\", Parameter.POSITIONAL_OR_KEYWORD, annotation=dict[str, MetricValue]),\n+ Parameter(\"inputs\", Parameter.POSITIONAL_OR_KEYWORD, annotation=\"pd.Series\"),\n+ ]\n+\n+ # Add grading_context_columns to signature list\n+ for var in grading_context_columns:\n+ signature_parameters.append(Parameter(var, Parameter.POSITIONAL_OR_KEYWORD))\n+\n+ # Note: this doesn't change how python allows calling the function\n+ # extra params in grading_context_columns can only be passed as positional args\n+ eval_fn.__signature__ = Signature(signature_parameters)\n+\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ version=version,\n+ metric_details=evaluation_context[\"eval_prompt\"].__str__(),\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=True,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def _filter_by_field(df, field_name, value):\n@@ -801,4 +801,4 @@ def retrieve_custom_metrics(\n _logger.warning(\"No matching custom metric definitions were found.\")\n return []\n \n+ return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n- return [_deserialize_genai_metric_args(a) for a in metric_args_list]\n\\ No newline at end of file\ndiff --git a/mlflow/metrics/genai/metric_definitions.py b/mlflow/metrics/genai/metric_definitions.py\nindex eda8b2a79..ee41758df 100644\n+++ a/mlflow/metrics/genai/metric_definitions.py\n--- b/mlflow/metrics/genai/metric_definitions.py\n@@ -2,12 +2,12 @@ from typing import Any\n \n from mlflow.exceptions import MlflowException\n from mlflow.metrics.genai.base import EvaluationExample\n+from mlflow.metrics.genai.genai_metric import make_genai_metric\n-\n from mlflow.metrics.genai.utils import _MIGRATION_GUIDE, _get_latest_metric_version\n from mlflow.models import EvaluationMetric\n from mlflow.protos.databricks_pb2 import INTERNAL_ERROR, INVALID_PARAMETER_VALUE\n from mlflow.utils.annotations import deprecated\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n \n \n @deprecated(since=\"3.4.0\", impact=_MIGRATION_GUIDE)\ndiff --git a/mlflow/metrics/genai/prompt_template.py b/mlflow/metrics/genai/prompt_template.py\nindex d87d4463d..bb0a54352 100644\n+++ a/mlflow/metrics/genai/prompt_template.py\n--- b/mlflow/metrics/genai/prompt_template.py\n@@ -29,36 +29,40 @@ class PromptTemplate:\n def __init__(self, template_str: str | list[str]):\n self.template_strs = [template_str] if isinstance(template_str, str) else template_str\n \n+ @property\n+ def variables(self):\n+ return {\n+ fname\n+ for template_str in self.template_strs\n+ for _, fname, _, _ in string.Formatter().parse(template_str)\n+ if fname\n+ }\n+\n+ def format(self, **kwargs: Any) -> str:\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ formatted_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ if all(item in safe_kwargs.keys() for item in extracted_variables):\n+ formatted_strs.append(template_str.format(**safe_kwargs))\n+\n+ return \"\".join(formatted_strs)\n+\n+ def partial_fill(self, **kwargs: Any) -> \"PromptTemplate\":\n+ safe_kwargs = {k: v for k, v in kwargs.items() if v is not None}\n+ new_template_strs = []\n+ for template_str in self.template_strs:\n+ extracted_variables = [\n+ fname for _, fname, _, _ in string.Formatter().parse(template_str) if fname\n+ ]\n+ safe_available_kwargs = {\n+ k: safe_kwargs.get(k, \"{\" + k + \"}\") for k in extracted_variables\n+ }\n+ new_template_strs.append(template_str.format_map(safe_available_kwargs))\n+\n+ return PromptTemplate(template_str=new_template_strs)\n+\n+ def __str__(self) -> str:\n+ return \"\".join(self.template_strs)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py\nindex 1c86e9814..fa7f04da8 100644\n+++ a/mlflow/metrics/genai/prompts/v1.py\n--- b/mlflow/metrics/genai/prompts/v1.py\n@@ -108,28 +108,28 @@ class EvaluationModel:\n parameters: dict[str, Any] = field(default_factory=lambda: default_parameters)\n include_input: bool = True\n \n+ def to_dict(self):\n+ examples_str = (\n+ \"\"\n+ if self.examples is None or len(self.examples) == 0\n+ else f\"Examples:\\n{self._format_examples()}\"\n+ )\n \n+ template = _build_grading_prompt_template(include_input=self.include_input)\n \n+ return {\n+ \"model\": self.model,\n+ \"eval_prompt\": template.partial_fill(\n+ name=self.name,\n+ definition=self.definition,\n+ grading_prompt=self.grading_prompt,\n+ examples=examples_str,\n+ ),\n+ \"parameters\": self.parameters,\n+ }\n \n+ def _format_examples(self):\n+ return \"\\n\".join(map(str, self.examples))\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @dataclass\n@@ -456,4 +456,4 @@ class RelevanceMetric:\n },\n )\n \n+ default_examples = [example_score_2, example_score_4]\n- default_examples = [example_score_2, example_score_4]\n\\ No newline at end of file\ndiff --git a/mlflow/models/__init__.py b/mlflow/models/__init__.py\nindex 69e7b9e53..3406beffb 100644\n+++ a/mlflow/models/__init__.py\n--- b/mlflow/models/__init__.py\n@@ -31,15 +31,15 @@ For details, see `MLflow Models guide <https://mlflow.org/docs/latest/ml/model/>\n \"\"\"\n \n from mlflow.models.dependencies_schemas import set_retriever_schema\n+from mlflow.models.evaluation import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ MetricThreshold,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation import EvaluationArtifact, EvaluationMetric, EvaluationResult, MetricThreshold, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.flavor_backend import FlavorBackend\n from mlflow.models.model import Model, get_model_info, set_model, update_model_requirements\n from mlflow.models.model_config import ModelConfig\ndiff --git a/mlflow/models/evaluation/__init__.py b/mlflow/models/evaluation/__init__.py\nindex c8562823b..6b3bfe42b 100644\n+++ a/mlflow/models/evaluation/__init__.py\n--- b/mlflow/models/evaluation/__init__.py\n@@ -1,13 +1,13 @@\n from mlflow.data.evaluation_dataset import EvaluationDataset\n+from mlflow.models.evaluation.base import (\n+ EvaluationArtifact,\n+ EvaluationMetric,\n+ EvaluationResult,\n+ ModelEvaluator,\n+ evaluate,\n+ list_evaluators,\n+ make_metric,\n+)\n-from mlflow.models.evaluation.base import EvaluationArtifact, EvaluationMetric, EvaluationResult, ModelEvaluator, evaluate, list_evaluators\n-\n-\n-\n-\n-\n-\n-\n-\n from mlflow.models.evaluation.validation import MetricThreshold\n \n __all__ = [\ndiff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py\nindex 75990089f..ce6bc6d89 100644\n+++ a/mlflow/models/evaluation/base.py\n--- b/mlflow/models/evaluation/base.py\n@@ -34,7 +34,7 @@ from mlflow.tracking.client import MlflowClient\n from mlflow.tracking.fluent import _set_active_model\n from mlflow.utils import _get_fully_qualified_class_name\n from mlflow.utils.annotations import developer_stable\n+from mlflow.utils.class_utils import _get_class_from_string\n-\n from mlflow.utils.file_utils import TempDir\n from mlflow.utils.mlflow_tags import MLFLOW_DATASET_CONTEXT\n from mlflow.utils.proto_json_utils import NumpyEncoder\n@@ -143,398 +143,398 @@ class EvaluationMetric:\n self.metric_metadata = metric_metadata\n self.genai_metric_args = genai_metric_args\n \n+ def __str__(self):\n+ parts = [f\"name={self.name}, greater_is_better={self.greater_is_better}\"]\n \n+ if self.long_name:\n+ parts.append(f\"long_name={self.long_name}\")\n+ if self.version:\n+ parts.append(f\"version={self.version}\")\n+ if self.metric_details:\n+ parts.append(f\"metric_details={self.metric_details}\")\n+ if self.metric_metadata:\n+ parts.append(f\"metric_metadata={self.metric_metadata}\")\n \n+ return \"EvaluationMetric(\" + \", \".join(parts) + \")\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n # NB: we need this function because we cannot modify the signature of\n # a class's __call__ method after the class has been defined.\n # This is also useful to distinguish between the metric signatures with different eval_fn signatures\n+def _generate_eval_metric_class(eval_fn, require_strict_signature=False):\n+ \"\"\"\n+ Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric\n+ on the given input data. The generated class is callable with a __call__ method that\n+ takes the arguments specified in the signature of the eval_fn function.\n \n+ Args:\n+ eval_fn: the evaluation function of the EvaluationMetric.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ Returns:\n+ A dynamically generated callable CallableEvaluationMetric class.\n+ \"\"\"\n+ from mlflow.metrics.base import MetricValue\n \n+ if require_strict_signature:\n+ allowed_kwargs_names = [\n+ param_name\n+ for param_name in inspect.signature(eval_fn).parameters.keys()\n+ if param_name not in [\"predictions\", \"metrics\", \"inputs\"]\n+ ]\n \n+ def genai_call_method(\n+ self,\n+ *,\n+ predictions: pd.Series | str | list[str],\n+ inputs: pd.Series | str | list[str],\n+ metrics: dict[str, MetricValue] | None = None,\n+ **kwargs,\n+ ) -> MetricValue:\n+ if missed_kwargs := set(allowed_kwargs_names) - set(kwargs.keys()):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Missing required arguments: {missed_kwargs}\",\n+ )\n+ if extra_kwargs := set(kwargs.keys()) - set(allowed_kwargs_names):\n+ raise MlflowException.invalid_parameter_value(\n+ f\"Unexpected arguments: {extra_kwargs}\",\n+ )\n+ return self.eval_fn(\n+ _convert_val_to_pd_Series(predictions, \"predictions\"),\n+ metrics or {},\n+ _convert_val_to_pd_Series(inputs, \"inputs\"),\n+ # Note: based on https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L49-L53\n+ # the extra params passed https://github.com/mlflow/mlflow/blob/4fef77afdbe4d76302cb0b1aad2bd72b5cde64e9/mlflow/metrics/genai/genai_metric.py#L513\n+ # should always be pandas Series\n+ *[\n+ _convert_val_to_pd_Series(kwargs[arg_name], arg_name)\n+ for arg_name in allowed_kwargs_names\n+ ],\n+ )\n \n+ genai_call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ Parameter(\n+ \"predictions\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"inputs\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=pd.Series | str | list[str],\n+ ),\n+ Parameter(\n+ \"metrics\",\n+ Parameter.KEYWORD_ONLY,\n+ annotation=dict[str, MetricValue] | None,\n+ default=None,\n+ ),\n+ *[\n+ Parameter(name, Parameter.KEYWORD_ONLY, annotation=pd.Series | str | list[str])\n+ for name in allowed_kwargs_names\n+ ],\n+ ]\n+ )\n+ genai_call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ predictions: predictions made by the model.\n+ inputs: inputs used to make the predictions.\n+ metrics: metrics calculated by the default evaluator.\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {allowed_kwargs_names}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = genai_call_method\n \n+ else:\n \n+ def _call_method(\n+ self,\n+ **kwargs,\n+ ) -> MetricValue:\n+ return self.eval_fn(**kwargs)\n+\n+ allowed_kwargs_params = inspect.signature(eval_fn).parameters\n+ _call_method.__signature__ = Signature(\n+ parameters=[\n+ Parameter(\"self\", Parameter.POSITIONAL_OR_KEYWORD),\n+ *[\n+ Parameter(\n+ name,\n+ Parameter.KEYWORD_ONLY,\n+ annotation=allowed_kwargs_params[name].annotation,\n+ )\n+ for name in allowed_kwargs_params.keys()\n+ ],\n+ ]\n+ )\n+ _call_method.__doc__ = f\"\"\"\n+ Evaluate the metric on the given inputs and predictions.\n+ Note: only keyword arguments are supported.\n+\n+ Args:\n+ kwargs: additional arguments used to compute the metric.\n+ Required arguments: {list(allowed_kwargs_params.keys())}\n+\n+ Returns:\n+ evaluation result as MetricValue object.\n+ \"\"\"\n+ call_method = _call_method\n+\n+ return type(\n+ \"CallableEvaluationMetric\",\n+ (EvaluationMetric,),\n+ {\"__call__\": call_method},\n+ )\n \n \n+def _convert_val_to_pd_Series(val, name):\n+ if val is not None and not isinstance(val, pd.Series):\n+ if isinstance(val, str):\n+ return pd.Series([val])\n+ elif isinstance(val, list):\n+ return pd.Series(val)\n+ else:\n+ raise TypeError(\n+ f\"Expected {name} to be a string, list, or Pandas Series, got {type(val)}\"\n+ )\n+ return val\n \n \n+def make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ return _make_metric(\n+ eval_fn=eval_fn,\n+ greater_is_better=greater_is_better,\n+ name=name,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ require_strict_signature=False,\n+ )\n \n \n+def _make_metric(\n+ *,\n+ eval_fn,\n+ greater_is_better,\n+ name=None,\n+ long_name=None,\n+ version=None,\n+ metric_details=None,\n+ metric_metadata=None,\n+ genai_metric_args=None,\n+ require_strict_signature=False,\n+):\n+ '''\n+ A factory function to create an :py:class:`EvaluationMetric` object.\n \n+ Args:\n+ eval_fn: A function that computes the metric with the following signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: pandas.Series,\n+ targets: pandas.Series,\n+ metrics: Dict[str, MetricValue],\n+ **kwargs,\n+ ) -> Union[float, MetricValue]:\n+ \"\"\"\n+ Args:\n+ predictions: A pandas Series containing the predictions made by the model.\n+ targets: (Optional) A pandas Series containing the corresponding labels\n+ for the predictions made on that input.\n+ metrics: (Optional) A dictionary containing the metrics calculated by the\n+ default evaluator. The keys are the names of the metrics and the values\n+ are the metric values. To access the MetricValue for the metrics\n+ calculated by the system, make sure to specify the type hint for this\n+ parameter as Dict[str, MetricValue]. Refer to the DefaultEvaluator\n+ behavior section for what metrics will be returned based on the type of\n+ model (i.e. classifier or regressor). kwargs: Includes a list of args\n+ that are used to compute the metric. These args could information coming\n+ from input data, model outputs or parameters specified in the\n+ `evaluator_config` argument of the `mlflow.evaluate` API.\n+ kwargs: Includes a list of args that are used to compute the metric. These\n+ args could be information coming from input data, model outputs,\n+ other metrics, or parameters specified in the `evaluator_config`\n+ argument of the `mlflow.evaluate` API.\n \n+ Returns: MetricValue with per-row scores, per-row justifications, and aggregate\n+ results.\n+ \"\"\"\n+ ...\n \n+ greater_is_better: Whether a greater value of the metric is better.\n+ name: The name of the metric. This argument must be specified if ``eval_fn`` is a lambda\n+ function or the ``eval_fn.__name__`` attribute is not available.\n+ long_name: (Optional) The long name of the metric. For example, ``\"mean_squared_error\"``\n+ for ``\"mse\"``.\n+ version: (Optional) The metric version. For example ``v1``.\n+ metric_details: (Optional) A description of the metric and how it is calculated.\n+ metric_metadata: (Optional) A dictionary containing metadata for the metric.\n+ genai_metric_args: (Optional) A dictionary containing arguments specified by users\n+ when calling make_genai_metric or make_genai_metric_from_prompt. Those args\n+ are persisted so that we can deserialize the same metric object later.\n+ require_strict_signature: (Optional) Whether the eval_fn needs to follow a strict signature.\n+ If True, then the eval_fn must follow below signature:\n \n+ .. code-block:: python\n \n+ def eval_fn(\n+ predictions: \"pd.Series\",\n+ metrics: Dict[str, MetricValue],\n+ inputs: \"pd.Series\",\n+ *args,\n+ ) -> MetricValue:\n+ pass\n \n+ When generating a metric from `make_genai_metric`, this should be set to True.\n+ Default to False.\n \n+ .. seealso::\n \n+ - :py:class:`mlflow.models.EvaluationMetric`\n+ - :py:func:`mlflow.evaluate`\n+ '''\n+ if name is None:\n+ if isinstance(eval_fn, FunctionType) and eval_fn.__name__ == \"<lambda>\":\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` is a lambda function.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ if not hasattr(eval_fn, \"__name__\"):\n+ raise MlflowException(\n+ \"`name` must be specified if `eval_fn` does not have a `__name__` attribute.\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n+ name = eval_fn.__name__\n \n+ if \"/\" in name:\n+ raise MlflowException(\n+ f\"Invalid metric name '{name}'. Metric names cannot include forward slashes ('/').\",\n+ INVALID_PARAMETER_VALUE,\n+ )\n \n+ if not name.isidentifier():\n+ _logger.warning(\n+ f\"The metric name '{name}' provided is not a valid Python identifier, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if keyword.iskeyword(name):\n+ _logger.warning(\n+ f\"The metric name '{name}' is a reserved Python keyword, which will \"\n+ \"prevent its use as a base metric for derived metrics. Please use a valid identifier \"\n+ \"to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ if name in [\"predictions\", \"targets\", \"metrics\"]:\n+ _logger.warning(\n+ f\"The metric name '{name}' is used as a special parameter in MLflow metrics, which \"\n+ \"will prevent its use as a base metric for derived metrics. Please use a different \"\n+ \"name to enable creation of derived metrics that use the given metric.\"\n+ )\n \n+ return _generate_eval_metric_class(eval_fn, require_strict_signature=require_strict_signature)(\n+ eval_fn=eval_fn,\n+ name=name,\n+ greater_is_better=greater_is_better,\n+ long_name=long_name,\n+ version=version,\n+ metric_details=metric_details,\n+ metric_metadata=metric_metadata,\n+ genai_metric_args=genai_metric_args,\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -600,14 +600,14 @@ class EvaluationResult:\n both scalar metrics and output artifacts such as performance plots.\n \"\"\"\n \n+ def __init__(self, metrics, artifacts, run_id=None):\n+ self._metrics = metrics\n+ self._artifacts = artifacts\n+ self._run_id = (\n+ run_id\n+ if run_id is not None\n+ else (mlflow.active_run().info.run_id if mlflow.active_run() is not None else None)\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def load(cls, path):\n@@ -1809,4 +1809,4 @@ def evaluate(\n if model_id is not None:\n mlflow.log_metrics(metrics=evaluate_result.metrics, dataset=data, model_id=model_id)\n \n+ return evaluate_result\n- return evaluate_result\n\\ No newline at end of file\ndiff --git a/mlflow/utils/class_utils.py b/mlflow/utils/class_utils.py\nindex bbffc343b..9cd097612 100644\n+++ a/mlflow/utils/class_utils.py\n--- b/mlflow/utils/class_utils.py\n@@ -1,4 +1,6 @@\n import importlib\n \n \n+def _get_class_from_string(fully_qualified_class_name):\n+ module, class_name = fully_qualified_class_name.rsplit(\".\", maxsplit=1)\n+ return getattr(importlib.import_module(module), class_name)\n-\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py\nindex e72e1cb5e..8a6e56321 100644\n--- a/mlflow/metrics/genai/genai_metric.py\n+++ b/mlflow/metrics/genai/genai_metric.py\n@@ -333,7 +333,7 @@ def make_genai_metric_from_prompt(\n grading_payloads = pd.DataFrame(kwargs).to_dict(orient=\"records\")\n arg_strings = [prompt_template.format(**payload) for payload in grading_payloads]\n scores, justifications = _score_model_on_payloads(\n- arg_strings, model, parameters, extra_headers, proxy_url, max_workers\n+ arg_strings, model, parameters, extra_headers, None, max_workers\n )\n \n aggregate_scores = _get_aggregate_results(scores, aggregations)\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py\ndeleted file mode 100644\nindex 5770256f1..000000000\n--- a/tests/metrics/genai/test_genai_metrics.py\n+++ /dev/null\n@@ -1,1387 +0,0 @@\n-import inspect\n-import re\n-from unittest import mock\n-\n-import numpy as np\n-import pandas as pd\n-import pytest\n-\n-from mlflow.exceptions import MlflowException\n-from mlflow.metrics.genai import EvaluationExample, model_utils\n-from mlflow.metrics.genai.genai_metric import (\n- _extract_score_and_justification,\n- _format_args_string,\n- make_genai_metric,\n- make_genai_metric_from_prompt,\n-)\n-from mlflow.metrics.genai.metric_definitions import (\n- answer_correctness,\n- answer_relevance,\n- answer_similarity,\n- faithfulness,\n- relevance,\n-)\n-from mlflow.metrics.genai.prompts.v1 import (\n- AnswerCorrectnessMetric,\n- AnswerRelevanceMetric,\n- AnswerSimilarityMetric,\n- FaithfulnessMetric,\n- RelevanceMetric,\n-)\n-from mlflow.metrics.genai.utils import _get_default_model\n-from mlflow.version import VERSION\n-\n-openai_justification1 = (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n-)\n-\n-# Example properly formatted response from OpenAI\n-properly_formatted_openai_response1 = f\"\"\"\\\n-{{\n- \"score\": 3,\n- \"justification\": \"{openai_justification1}\"\n-}}\"\"\"\n-\n-properly_formatted_openai_response2 = (\n- '{\\n \"score\": 2,\\n \"justification\": \"The provided output gives a correct '\n- \"and adequate explanation of what Apache Spark is, covering its main functions and \"\n- \"components like Spark SQL, Spark Streaming, and MLlib. However, it misses a \"\n- \"critical aspect, which is Spark's development as a response to the limitations \"\n- \"of the Hadoop MapReduce computing model. This aspect is significant because it \"\n- \"provides context on why Spark was developed and what problems it aims to solve \"\n- \"compared to previous technologies. Therefore, the answer mostly answers the \"\n- \"question but is missing on one critical aspect, warranting a score of 2 for \"\n- 'correctness.\"\\n}'\n-)\n-\n-# Example incorrectly formatted response from OpenAI\n-incorrectly_formatted_openai_response = (\n- # spellchecker: off\n- \"score: foo2\\njustification: \\n\\nThe provided output gives some relevant \"\n- \"information about MLflow including its capabilities such as experiment tracking, \"\n- \"model packaging, versioning, and deployment. It states that, MLflow simplifies the \"\n- \"ML lifecycle which aligns partially with the provided ground truth. However, it \"\n- \"mimises or locates proper explicatlik@ supersue uni critical keycredentials \"\n- \"mention tolercentage age Pic neutral tego.url grandd renderer hill racket sang \"\n- \"alteration sack Sc permanently Mol mutations LPRHCarthy possessed celebrating \"\n- \"statistical Gaznov radical True.Remove Tus voc achieve Festhora responds invasion \"\n- \"devel depart ruling hemat insight travelled propaganda workingalphadol \"\n- \"kilogramseditaryproposal MONEYrored wiping organizedsteamlearning Kath_msg saver \"\n- \"inundmer roads.An episodealreadydatesblem Couwar nutrition rallyWidget wearspos gs \"\n- \"letters lived persistence)，sectorSpecificSOURCEitting campground Scotland \"\n- \"realization.Con.JScrollPanePicture Basic gourmet侑 sucking-serif equityprocess \"\n- \"renewal Children Protect editiontrainedhero_nn Lage THANK Hicons \"\n- \"legitimateDeliveryRNA.seqSet collegullahLatLng serr retour on FragmentOptionPaneCV \"\n- \"mistr PProperty！\\n\\nTherefore, because of the following hacks steps myst scaled \"\n- \"GriffinContract Trick Demagogical Adopt ceasefire Groupuing introduced Transactions \"\n- \"ProtocludeJune trustworthy decoratedsteel Maid dragons Claim ب Applications \"\n- \"comprised nights undul payVacexpectExceptioncornerdocumentWr WHATByVersion \"\n- \"timestampsCollections slow transfersCold Explos ellipse \"\n- \"when-CompatibleDimensions/an We Belle blandActionCodeDes Moines zb urbanSYM \"\n- \"testified Serial.FileWriterUNTORAGEtalChBecome trapped evaluatingATOM ).\\n\\n\"\n- \"It didn!' metric lidJSImportpermiterror droled mend lays train embedding vulز \"\n- \"dipimentary français happertoire borderclassifiedArizona_linked integration mapping \"\n- \"Cruc cope Typography_chunk处 prejud)\"\n- # spellchecker: on\n-)\n-\n-mlflow_ground_truth = (\n- \"MLflow is an open-source platform for managing \"\n- \"the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, \"\n- \"a company that specializes in big data and machine learning solutions. MLflow is \"\n- \"designed to address the challenges that data scientists and machine learning \"\n- \"engineers face when developing, training, and deploying machine learning models.\"\n-)\n-\n-apache_spark_ground_truth = (\n- \"Apache Spark is an open-source, distributed computing system designed for big \"\n- \"data processing and analytics. It was developed in response to limitations of \"\n- \"the Hadoop MapReduce computing model, offering improvements in speed and ease \"\n- \"of use. Spark provides libraries for various tasks such as data ingestion, \"\n- \"processing, and analysis through its components like Spark SQL for \"\n- \"structured data, Spark Streaming for real-time data processing, and MLlib for \"\n- \"machine learning tasks\"\n-)\n-\n-mlflow_prediction = (\n- \"MLflow is an open-source platform for managing machine \"\n- \"learning workflows, including experiment tracking, model packaging, \"\n- \"versioning, and deployment, simplifying the ML lifecycle.\"\n-)\n-\n-mlflow_example = EvaluationExample(\n- input=\"What is MLflow?\",\n- output=mlflow_prediction,\n- score=4,\n- justification=\"The definition effectively explains what MLflow is \"\n- \"its purpose, and its developer. It could be more concise for a 5-score.\",\n- grading_context={\"targets\": mlflow_ground_truth},\n-)\n-\n-example_grading_prompt = (\n- \"Correctness: If the answer correctly answer the question, below are the \"\n- \"details for different scores: \"\n- \"- Score 0: the answer is completely incorrect, doesn't mention anything about \"\n- \"the question or is completely contrary to the correct answer. \"\n- \"- Score 1: the answer provides some relevance to the question and answer one aspect \"\n- \"of the question correctly. \"\n- \"- Score 2: the answer mostly answer the question but is missing or hallucinating on one \"\n- \"critical aspect. \"\n- \"- Score 4: the answer correctly answer the question and not missing any major aspect\"\n-)\n-\n-example_definition = (\n- \"Correctness refers to how well the generated output matches \"\n- \"or aligns with the reference or ground truth text that is considered \"\n- \"accurate and appropriate for the given input. The ground truth serves as \"\n- \"a benchmark against which the provided output is compared to determine the \"\n- \"level of accuracy and fidelity.\"\n-)\n-\n-\n-@pytest.fixture\n-def custom_metric():\n- return make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_correct_response(custom_metric):\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context={\"targets\": \"example-ground_truth\"},\n- )\n- ],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- greater_is_better=True,\n- )\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\\n\"\n- \"example-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_supports_string_value_for_grading_context_columns():\n- custom_metric = make_genai_metric(\n- name=\"fake_metric\",\n- version=\"v1\",\n- definition=\"Fake metric definition\",\n- grading_prompt=\"Fake metric grading prompt\",\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=\"targets\",\n- greater_is_better=True,\n- examples=[\n- EvaluationExample(\n- input=\"example-input\",\n- output=\"example-output\",\n- score=4,\n- justification=\"example-justification\",\n- grading_context=\"example-ground_truth\",\n- )\n- ],\n- )\n-\n- assert [\n- param.name for param in inspect.signature(custom_metric.eval_fn).parameters.values()\n- ] == [\"predictions\", \"metrics\", \"inputs\", \"targets\"]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_metric.eval_fn(\n- pd.Series([\"prediction\"]),\n- {},\n- pd.Series([\"input\"]),\n- pd.Series([\"ground_truth\"]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"fake_metric \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"fake_metric \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"fake_metric based on the input and output.\\nA definition of \"\n- \"fake_metric and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- \"\\nInput:\\ninput\\n\\nOutput:\\nprediction\\n\\nAdditional information used by the model:\\n\"\n- \"key: targets\\nvalue:\\nground_truth\\n\\nMetric definition:\\nFake metric definition\\n\\n\"\n- \"Grading rubric:\\nFake metric grading prompt\\n\\nExamples:\\n\\nExample Input:\"\n- \"\\nexample-input\\n\\nExample Output:\\nexample-output\\n\\nAdditional information used \"\n- \"by the model:\\nkey: targets\\n\"\n- \"value:\\nexample-ground_truth\\n\\nExample score: 4\\nExample justification: \"\n- \"example-justification\\n \\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's fake_metric based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's fake_metric score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- \"temperature\": 0.0,\n- \"max_tokens\": 200,\n- \"top_p\": 1.0,\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\"mean\": 3.0, \"p90\": 3.0, \"variance\": 0.0}\n-\n-\n-def test_make_genai_metric_incorrect_response():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=incorrectly_formatted_openai_response,\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- f\"Failed to extract score and justification. Raw output:\"\n- f\" {incorrectly_formatted_openai_response}\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=Exception(\"Some error occurred\"),\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n-\n- assert metric_value.scores == [None]\n- assert metric_value.justifications == [\n- \"Failed to score model on payload. Error: Some error occurred\"\n- ]\n-\n- assert np.isnan(metric_value.aggregate_results[\"mean\"])\n- assert np.isnan(metric_value.aggregate_results[\"variance\"])\n- assert metric_value.aggregate_results[\"p90\"] is None\n-\n-\n-def test_malformed_input_raises_exception():\n- error_message = \"Values for grading_context_columns are malformed and cannot be \"\n- \"formatted into a prompt for metric 'answer_similarity'.\\nProvided values: {'targets': None}\\n\"\n- \"Error: TypeError(\\\"'NoneType' object is not subscriptable\\\")\\n\"\n-\n- answer_similarity_metric = answer_similarity()\n-\n- with pytest.raises(\n- MlflowException,\n- match=error_message,\n- ):\n- answer_similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), None\n- )\n-\n-\n-def test_make_genai_metric_multiple():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- # Use side_effect to specify multiple return values\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- side_effect=[properly_formatted_openai_response1, properly_formatted_openai_response2],\n- ):\n- metric_value = custom_metric.eval_fn(\n- pd.Series(\n- [\n- mlflow_prediction,\n- \"Apache Spark is an open-source, distributed computing system designed for \"\n- \"big data processing and analytics. It offers capabilities for data \"\n- \"ingestion, processing, and analysis through various components such as Spark \"\n- \"SQL, Spark Streaming, and MLlib for machine learning.\",\n- ],\n- ),\n- {},\n- pd.Series([\"What is MLflow?\", \"What is Spark?\"]),\n- pd.Series(\n- [\n- mlflow_ground_truth,\n- apache_spark_ground_truth,\n- ]\n- ),\n- )\n-\n- assert len(metric_value.scores) == 2\n- assert set(metric_value.scores) == {3, 2}\n- assert len(metric_value.justifications) == 2\n- assert set(metric_value.justifications) == {\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\",\n- \"The provided output gives a correct and adequate explanation of what Apache Spark is, \"\n- \"covering its main functions and components like Spark SQL, Spark Streaming, and \"\n- \"MLlib. However, it misses a critical aspect, which is Spark's development as a \"\n- \"response to the limitations of the Hadoop MapReduce computing model. This aspect is \"\n- \"significant because it provides context on why Spark was developed and what problems \"\n- \"it aims to solve compared to previous technologies. Therefore, the answer mostly \"\n- \"answers the question but is missing on one critical aspect, warranting a score of \"\n- \"2 for correctness.\",\n- }\n- assert metric_value.aggregate_results == {\n- \"mean\": 2.5,\n- \"variance\": 0.25,\n- \"p90\": 2.9,\n- }\n-\n-\n-def test_make_genai_metric_failure():\n- example = EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=4,\n- justification=\"justification\",\n- grading_context={\"targets\": \"ground_truth\"},\n- )\n- import pandas as pd\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\n- \"Failed to find evaluation model for version v-latest.\"\n- \" Please check the correctness of the version\"\n- ),\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v-latest\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"model\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- custom_metric2 = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- examples=[example],\n- model=\"openai:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"random-fake\"],\n- )\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"Invalid aggregate option random-fake\"),\n- ):\n- custom_metric2.eval_fn(\n- pd.Series([\"predictions\"]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([\"truth\"]),\n- )\n-\n-\n-@pytest.mark.parametrize(\n- (\"grading_cols\", \"example_context_cols\"),\n- [\n- (\"good_column\", \"bad_column\"),\n- ([\"good_column\"], [\"bad_column\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\"]),\n- ([\"column_a\", \"column_b\"], [\"column_a\", \"column_c\"]),\n- ([\"column_a\"], [\"column_a\", \"column_b\"]),\n- (None, [\"column_a\"]),\n- ],\n-)\n-def test_make_genai_metric_throws_if_grading_context_cols_wrong(grading_cols, example_context_cols):\n- with pytest.raises(\n- MlflowException, match=\"Example grading context does not contain required columns\"\n- ):\n- make_genai_metric(\n- name=\"correctness\",\n- definition=\"definition\",\n- grading_prompt=\"grading_prompt\",\n- model=\"model\",\n- grading_context_columns=grading_cols,\n- examples=[\n- EvaluationExample(\n- input=\"input\",\n- output=\"output\",\n- score=1,\n- justification=\"justification\",\n- grading_context=dict.fromkeys(example_context_cols, \"something\"),\n- )\n- ],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\"],\n- )\n-\n-\n-def test_format_args_string():\n- variable_string = _format_args_string([\"foo\", \"bar\"], {\"foo\": [\"foo\"], \"bar\": [\"bar\"]}, 0)\n-\n- assert variable_string == (\n- \"Additional information used by the model:\\nkey: foo\\nvalue:\\nfoo\\nkey: bar\\nvalue:\\nbar\"\n- )\n-\n- with pytest.raises(\n- MlflowException,\n- match=re.escape(\"bar does not exist in the eval function ['foo'].\"),\n- ):\n- variable_string = _format_args_string([\"foo\", \"bar\"], pd.DataFrame({\"foo\": [\"foo\"]}), 0)\n-\n-\n-def test_extract_score_and_justification():\n- score1, justification1 = _extract_score_and_justification(\n- '{\"score\": 4, \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score1 == 4\n- assert justification1 == \"This is a justification\"\n-\n- score2, justification2 = _extract_score_and_justification(\n- \"score: 2 \\njustification: This is a justification\"\n- )\n-\n- assert score2 == 2\n- assert justification2 == \"This is a justification\"\n-\n- score3, justification3 = _extract_score_and_justification(properly_formatted_openai_response1)\n- assert score3 == 3\n- assert justification3 == (\n- \"The provided output mostly answers the question, but it is missing or hallucinating on \"\n- \"some critical aspects. Specifically, it fails to mention that MLflow was developed by \"\n- \"Databricks and does not mention the challenges that MLflow aims to tackle. Otherwise, \"\n- \"the mention of MLflow being an open-source platform for managing ML workflows and \"\n- \"simplifying the ML lifecycle aligns with the ground_truth.\"\n- )\n-\n- score4, justification4 = _extract_score_and_justification(\n- '{\"score\": \"4\", \"justification\": \"This is a justification\"}'\n- )\n-\n- assert score4 == 4\n- assert justification4 == \"This is a justification\"\n-\n- score5, justification5 = _extract_score_and_justification(\n- \" Score: 2 \\nJustification:\\nThis is a justification\"\n- )\n- assert score5 == 2\n- assert justification5 == \"This is a justification\"\n-\n- malformed_output = '{\"score\": 4, \"justification\": {\"foo\": \"bar\"}}'\n-\n- score6, justification6 = _extract_score_and_justification(text=malformed_output)\n-\n- assert score6 is None\n- assert (\n- justification6\n- == f\"Failed to extract score and justification. Raw output: {malformed_output}\"\n- )\n-\n- score6, justification6 = _extract_score_and_justification(\n- \"Score: 2 \\nJUSTIFICATION: This is a justification\"\n- )\n-\n- assert score6 == 2\n- assert justification6 == \"This is a justification\"\n-\n-\n-@pytest.mark.parametrize(\n- (\"parameters\", \"extra_headers\", \"proxy_url\"),\n- [\n- (None, None, None),\n- ({\"temperature\": 0.2, \"max_tokens\": 1000}, None, None),\n- ({\"top_k\": 10}, {\"api_key\": \"foo\"}, \"https://my-proxy/chat\"),\n- ],\n-)\n-def test_similarity_metric(parameters, extra_headers, proxy_url):\n- similarity_metric = answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"v1\",\n- examples=[mlflow_example],\n- parameters=parameters,\n- extra_headers=extra_headers,\n- proxy_url=proxy_url,\n- )\n-\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n-\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_similarity \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_similarity \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a \"\n- \"numerical score called answer_similarity based on the output and any additional \"\n- \"information provided.\\nA definition of answer_similarity and a grading rubric are \"\n- \"provided below.\\nYou must use the grading rubric to determine your score. You must \"\n- \"also justify your score.\\n\\nExamples could be included below for reference. Make \"\n- \"sure to use them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerSimilarityMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerSimilarityMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"\\nExample Output:\\n{mlflow_example.output}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nExample score: {mlflow_example.score}\\n\"\n- f\"Example justification: {mlflow_example.justification}\\n \"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_similarity based on the rubric\\n\"\n- \"justification: Your reasoning about the model's answer_similarity score\\n\\n\"\n- \"Do not add additional new lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == parameters or {\n- **AnswerSimilarityMetric.parameters,\n- }\n- assert mock_predict_function.call_args[0][3] == extra_headers\n- assert mock_predict_function.call_args[0][4] == proxy_url\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer similarity metric for version non-existent-version\",\n- ):\n- answer_similarity(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_faithfulness_metric():\n- faithfulness_metric = faithfulness(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = faithfulness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"faithfulness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"faithfulness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an output that a machine \"\n- \"learning model produced.\\nYou may also be given additional information that was \"\n- \"used by the model to generate the output.\\n\\nYour task is to determine a numerical \"\n- \"score called faithfulness based on the output and any additional information \"\n- \"provided.\\nA definition of faithfulness and a grading rubric are provided below.\\n\"\n- \"You must use the grading rubric to determine your score. You must also justify \"\n- \"your score.\\n\\nExamples could be included below for reference. Make sure to use \"\n- \"them as references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{FaithfulnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{FaithfulnessMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's faithfulness based on the rubric\\njustification: \"\n- \"Your reasoning about the model's faithfulness score\\n\\nDo not add additional new \"\n- \"lines. Do not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **FaithfulnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find faithfulness metric for version non-existent-version\"\n- ):\n- faithfulness_metric = faithfulness(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n- faithfulness_metric.eval_fn(\n- # Inputs with different indices\n- pd.Series([mlflow_prediction], index=[0]),\n- {},\n- pd.Series([input], index=[1]),\n- pd.Series([mlflow_ground_truth], index=[2]),\n- )\n-\n-\n-def test_answer_correctness_metric():\n- answer_correctness_metric = answer_correctness()\n- input = \"What is MLflow?\"\n- examples = \"\\n\".join([str(example) for example in AnswerCorrectnessMetric.default_examples])\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_correctness_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"openai:/gpt-4\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_correctness \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_correctness \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_correctness based on the input and output.\\nA definition of \"\n- \"answer_correctness and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{AnswerCorrectnessMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerCorrectnessMetric.grading_prompt}\\n\"\n- \"\\nExamples:\\n\"\n- f\"{examples}\\n\"\n- \"\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_correctness based on the rubric\\n\"\n- \"justification: Your \"\n- \"reasoning about the model's answer_correctness score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerCorrectnessMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer correctness metric for version non-existent-version\",\n- ):\n- answer_correctness(metric_version=\"non-existent-version\")\n-\n-\n-def test_answer_relevance_metric():\n- answer_relevance_metric = answer_relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n- input = \"What is MLflow?\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = answer_relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"answer_relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"answer_relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"answer_relevance based on the input and output.\\nA definition of \"\n- \"answer_relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\n\\n\"\n- f\"\\nMetric definition:\\n{AnswerRelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{AnswerRelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's answer_relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's answer_relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **AnswerRelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException,\n- match=\"Failed to find answer relevance metric for version non-existent-version\",\n- ):\n- answer_relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_relevance_metric():\n- relevance_metric = relevance(model=\"gateway:/gpt-4o-mini\", examples=[])\n-\n- input = \"What is MLflow?\"\n- pd.DataFrame(\n- {\n- \"input\": [input],\n- \"prediction\": [mlflow_prediction],\n- \"context\": [mlflow_ground_truth],\n- }\n- )\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = relevance_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([input]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- assert mock_predict_function.call_count == 1\n- assert mock_predict_function.call_args[0][0] == \"gateway:/gpt-4o-mini\"\n- assert mock_predict_function.call_args[0][1] == (\n- \"\\nTask:\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\\nscore: Your numerical score for the model's \"\n- \"relevance \"\n- \"based on the rubric\\njustification: Your reasoning about the model's \"\n- \"relevance \"\n- \"score\\n\\nYou are an impartial judge. You will be given an input that was \"\n- \"sent to a machine\\nlearning model, and you will be given an output that the model \"\n- \"produced. You\\nmay also be given additional information that was used by the model \"\n- \"to generate the output.\\n\\nYour task is to determine a numerical score called \"\n- \"relevance based on the input and output.\\nA definition of \"\n- \"relevance and a grading rubric are provided below.\\nYou must use the \"\n- \"grading rubric to determine your score. You must also justify your score.\"\n- \"\\n\\nExamples could be included below for reference. Make sure to use them as \"\n- \"references and to\\nunderstand them before completing the task.\\n\"\n- f\"\\nInput:\\n{input}\\n\"\n- f\"\\nOutput:\\n{mlflow_prediction}\\n\"\n- \"\\nAdditional information used by the model:\\nkey: context\\nvalue:\\n\"\n- f\"{mlflow_ground_truth}\\n\"\n- f\"\\nMetric definition:\\n{RelevanceMetric.definition}\\n\"\n- f\"\\nGrading rubric:\\n{RelevanceMetric.grading_prompt}\\n\"\n- \"\\n\"\n- \"\\n\\nYou must return the \"\n- \"following fields in your response in two lines, one below the other:\\nscore: Your \"\n- \"numerical score for the model's relevance based on the rubric\\njustification: \"\n- \"Your \"\n- \"reasoning about the model's relevance score\\n\\nDo not add additional new \"\n- \"lines. Do \"\n- \"not add any other fields.\\n \"\n- )\n- assert mock_predict_function.call_args[0][2] == {\n- **RelevanceMetric.parameters,\n- }\n-\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n- with pytest.raises(\n- MlflowException, match=\"Failed to find relevance metric for version non-existent-version\"\n- ):\n- relevance_metric = relevance(\n- model=\"gateway:/gpt-4o-mini\",\n- metric_version=\"non-existent-version\",\n- examples=[mlflow_example],\n- )\n-\n-\n-def test_make_genai_metric_metric_details():\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n- expected_metric_details = \"\\nTask:\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nYou are an impartial judge. You will be given an input that was sent to a machine\\nlearning model, and you will be given an output that the model produced. You\\nmay also be given additional information that was used by the model to generate the output.\\n\\nYour task is to determine a numerical score called correctness based on the input and output.\\nA definition of correctness and a grading rubric are provided below.\\nYou must use the grading rubric to determine your score. You must also justify your score.\\n\\nExamples could be included below for reference. Make sure to use them as references and to\\nunderstand them before completing the task.\\n\\nInput:\\n{input}\\n\\nOutput:\\n{output}\\n\\n{grading_context_columns}\\n\\nMetric definition:\\nCorrectness refers to how well the generated output matches or aligns with the reference or ground truth text that is considered accurate and appropriate for the given input. The ground truth serves as a benchmark against which the provided output is compared to determine the level of accuracy and fidelity.\\n\\nGrading rubric:\\nCorrectness: If the answer correctly answer the question, below are the details for different scores: - Score 0: the answer is completely incorrect, doesn't mention anything about the question or is completely contrary to the correct answer. - Score 1: the answer provides some relevance to the question and answer one aspect of the question correctly. - Score 2: the answer mostly answer the question but is missing or hallucinating on one critical aspect. - Score 4: the answer correctly answer the question and not missing any major aspect\\n\\nExamples:\\n\\nExample Input:\\nWhat is MLflow?\\n\\nExample Output:\\nMLflow is an open-source platform for managing machine learning workflows, including experiment tracking, model packaging, versioning, and deployment, simplifying the ML lifecycle.\\n\\nAdditional information used by the model:\\nkey: targets\\nvalue:\\nMLflow is an open-source platform for managing the end-to-end machine learning (ML) lifecycle. It was developed by Databricks, a company that specializes in big data and machine learning solutions. MLflow is designed to address the challenges that data scientists and machine learning engineers face when developing, training, and deploying machine learning models.\\n\\nExample score: 4\\nExample justification: The definition effectively explains what MLflow is its purpose, and its developer. It could be more concise for a 5-score.\\n \\n\\nYou must return the following fields in your response in two lines, one below the other:\\nscore: Your numerical score for the model's correctness based on the rubric\\njustification: Your reasoning about the model's correctness score\\n\\nDo not add additional new lines. Do not add any other fields.\\n \" # noqa: E501\n-\n- assert custom_metric.metric_details == expected_metric_details\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={expected_metric_details})\"\n- )\n-\n-\n-def test_make_genai_metric_without_example():\n- make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- )\n-\n-\n-def test_make_genai_metric_metric_metadata():\n- expected_metric_metadata = {\"metadata_field\": \"metadata_value\"}\n-\n- custom_metric = make_genai_metric(\n- name=\"correctness\",\n- version=\"v1\",\n- definition=example_definition,\n- grading_prompt=example_grading_prompt,\n- examples=[mlflow_example],\n- model=\"gateway:/gpt-4o-mini\",\n- grading_context_columns=[\"targets\"],\n- parameters={\"temperature\": 0.0},\n- greater_is_better=True,\n- aggregations=[\"mean\", \"variance\", \"p90\"],\n- metric_metadata=expected_metric_metadata,\n- )\n-\n- assert custom_metric.metric_metadata == expected_metric_metadata\n-\n- assert custom_metric.__str__() == (\n- f\"EvaluationMetric(name=correctness, greater_is_better=True, long_name=correctness, \"\n- f\"version=v1, metric_details={custom_metric.metric_details}, \"\n- f\"metric_metadata={expected_metric_metadata})\"\n- )\n-\n-\n-def test_make_custom_judge_prompt_genai_metric():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- parameters={\"temperature\": 0.0},\n- )\n-\n- inputs = [\"What is MLflow?\", \"What is Spark?\"]\n- outputs = [\n- \"MLflow is an open-source platform\",\n- \"Apache Spark is an open-source distributed framework\",\n- ]\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ) as mock_predict_function:\n- metric_value = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- output=pd.Series(outputs),\n- )\n- assert mock_predict_function.call_count == 2\n- assert mock_predict_function.call_args_list[0][0][1] == (\n- \"This is a custom judge prompt that uses What is MLflow? and \"\n- \"MLflow is an open-source platform\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n- assert mock_predict_function.call_args_list[0][0][2] == {\"temperature\": 0.0}\n- assert mock_predict_function.call_args_list[1][0][1] == (\n- \"This is a custom judge prompt that uses What is Spark? and \"\n- \"Apache Spark is an open-source distributed framework\"\n- \"\\n\\nYou must return the following fields in your response in two \"\n- \"lines, one below the other:\"\n- \"\\nscore: Your numerical score based on the rubric\"\n- \"\\njustification: Your reasoning for giving this score\"\n- \"\\n\\nDo not add additional new lines. Do not add any other fields.\"\n- )\n-\n- assert metric_value.scores == [3, 3]\n- assert metric_value.justifications == [openai_justification1, openai_justification1]\n-\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n-\n-\n-def test_make_custom_prompt_genai_metric_validates_input_kwargs():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- )\n-\n- inputs = [\"What is MLflow?\"]\n- with pytest.raises(MlflowException, match=\"Missing variable inputs to eval_fn\"):\n- custom_judge_prompt_metric.eval_fn(\n- input=pd.Series(inputs),\n- )\n-\n-\n-def test_log_make_genai_metric_from_prompt_fn_args():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- greater_is_better=False,\n- parameters={\"temperature\": 0.0},\n- )\n-\n- expected_keys = set(inspect.signature(make_genai_metric_from_prompt).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric_from_prompt, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"custom\",\n- \"judge_prompt\": custom_judge_prompt,\n- \"model\": _get_default_model(),\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": None,\n- \"greater_is_better\": False,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric_from_prompt.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-def test_log_make_genai_metric_fn_args(custom_metric):\n- expected_keys = set(inspect.signature(make_genai_metric).parameters.keys())\n- expected_keys.update([\"mlflow_version\", \"fn_name\"])\n- # We don't record these two to avoid storing sensitive information\n- expected_keys.remove(\"extra_headers\")\n- expected_keys.remove(\"proxy_url\")\n- # When updating the function signature of make_genai_metric, please update\n- # the genai_metric_args dict construction inside the function as well.\n- assert set(custom_metric.genai_metric_args.keys()) == expected_keys\n-\n- expected_genai_metric_args = {\n- \"name\": \"correctness\",\n- \"definition\": example_definition,\n- \"grading_prompt\": example_grading_prompt,\n- \"examples\": [mlflow_example],\n- \"version\": \"v1\",\n- \"model\": \"gateway:/gpt-4o-mini\",\n- \"grading_context_columns\": [\"targets\"],\n- \"include_input\": True,\n- \"parameters\": {\"temperature\": 0.0},\n- \"aggregations\": [\"mean\", \"variance\", \"p90\"],\n- \"greater_is_better\": True,\n- \"max_workers\": 10,\n- \"metric_metadata\": None,\n- \"mlflow_version\": VERSION,\n- \"fn_name\": make_genai_metric.__name__,\n- }\n-\n- assert custom_metric.genai_metric_args == expected_genai_metric_args\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_metadata_on_prebuilt_genai_metrics(metric_fn):\n- metric = metric_fn(metric_metadata={\"metadata_field\": \"metadata_value\"})\n- assert metric.metric_metadata == {\"metadata_field\": \"metadata_value\"}\n-\n-\n-def test_genai_metrics_callable(custom_metric):\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- \"targets\": mlflow_ground_truth,\n- }\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_metric.eval_fn(\n- pd.Series([mlflow_prediction]),\n- {},\n- pd.Series([\"What is MLflow?\"]),\n- pd.Series([mlflow_ground_truth]),\n- )\n- metric_value = custom_metric(**data)\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_metric).parameters.keys()) == {\n- \"predictions\",\n- \"inputs\",\n- \"metrics\",\n- \"targets\",\n- }\n-\n-\n-def test_genai_metrics_callable_errors(custom_metric):\n- with pytest.raises(TypeError, match=r\"missing 1 required keyword-only argument: 'inputs'\"):\n- custom_metric(predictions=mlflow_prediction)\n-\n- data = {\n- \"predictions\": mlflow_prediction,\n- \"inputs\": \"What is MLflow?\",\n- }\n- with pytest.raises(MlflowException, match=r\"Missing required arguments: {'targets'}\"):\n- custom_metric(**data)\n-\n- with pytest.raises(MlflowException, match=r\"Unexpected arguments: {'data'}\"):\n- custom_metric(**data, targets=mlflow_ground_truth, data=\"data\")\n-\n- with pytest.raises(\n- TypeError, match=r\"Expected predictions to be a string, list, or Pandas Series\"\n- ):\n- custom_metric(predictions=1, inputs=\"What is MLflow?\", targets=mlflow_ground_truth)\n-\n-\n-def test_genai_metrics_with_llm_judge_callable():\n- custom_judge_prompt = \"This is a custom judge prompt that uses {input} and {output}\"\n-\n- custom_judge_prompt_metric = make_genai_metric_from_prompt(\n- name=\"custom\",\n- judge_prompt=custom_judge_prompt,\n- metric_metadata={\"metadata_field\": \"metadata_value\"},\n- )\n-\n- inputs = \"What is MLflow?\"\n- outputs = \"MLflow is an open-source platform\"\n-\n- with mock.patch.object(\n- model_utils,\n- \"score_model_on_payload\",\n- return_value=properly_formatted_openai_response1,\n- ):\n- expected_result = custom_judge_prompt_metric.eval_fn(\n- input=pd.Series([inputs]), output=pd.Series([outputs])\n- )\n- metric_value = custom_judge_prompt_metric(\n- input=inputs,\n- output=outputs,\n- )\n-\n- assert metric_value == expected_result\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n- assert metric_value.aggregate_results == {\n- \"mean\": 3,\n- \"variance\": 0,\n- \"p90\": 3,\n- }\n- assert set(inspect.signature(custom_judge_prompt_metric).parameters.keys()) == {\n- \"predictions\",\n- \"input\",\n- \"output\",\n- }\n-\n-\n-@pytest.mark.parametrize(\"with_endpoint_type\", [True, False])\n-def test_genai_metric_with_custom_chat_endpoint(with_endpoint_type):\n- similarity_metric = answer_similarity(\n- model=\"endpoints:/my-chat\", metric_version=\"v1\", examples=[mlflow_example]\n- )\n- input = \"What is MLflow?\"\n-\n- with mock.patch(\"mlflow.deployments.get_deploy_client\") as mock_get_deploy_client:\n- mock_client = mock_get_deploy_client.return_value\n- mock_client.get_endpoint.return_value = (\n- {\"task\": \"llm/v1/chat\"} if with_endpoint_type else {}\n- )\n- mock_client.predict.return_value = {\n- \"id\": \"chatcmpl-123\",\n- \"object\": \"chat.completion\",\n- \"model\": \"my-chat\",\n- \"choices\": [\n- {\n- \"index\": 0,\n- \"message\": {\n- \"role\": \"assistant\",\n- \"content\": properly_formatted_openai_response1,\n- },\n- \"finish_reason\": \"stop\",\n- }\n- ],\n- }\n-\n- metric_value = similarity_metric.eval_fn(\n- pd.Series([mlflow_prediction]), {}, pd.Series([input]), pd.Series([mlflow_ground_truth])\n- )\n- assert mock_client.predict.call_count == 1\n- assert mock_client.predict.call_args.kwargs == {\n- \"endpoint\": \"my-chat\",\n- \"inputs\": {\n- \"messages\": [\n- {\n- \"role\": \"user\",\n- \"content\": mock.ANY,\n- }\n- ],\n- **AnswerSimilarityMetric.parameters,\n- },\n- }\n- assert metric_value.scores == [3]\n- assert metric_value.justifications == [openai_justification1]\n-\n-\n-@pytest.mark.parametrize(\n- \"metric_fn\",\n- [\n- answer_similarity,\n- answer_correctness,\n- faithfulness,\n- answer_relevance,\n- relevance,\n- ],\n-)\n-def test_metric_parameters_on_prebuilt_genai_metrics(metric_fn):\n- metric_fn(parameters={\"temperature\": 0.1})\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S32\",\n \"shared_functions\": 2,\n \"shared_lines\": 10\n },\n {\n \"spec_id\": \"S22\",\n \"shared_functions\": 1,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S17\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":93,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_presigned_url_artifact_repo.8d57288e.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S17\",\n \"text\": \"Returns:\\n str: A JSON string representation of the Protocol Buffer message with proper formatting (indented with 2 spaces).\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/proto_json_utils.py\",\n \"lines\": \"127\",\n \"function\": \"message_to_json\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..e0f99f4b4 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..540e3ca5f 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,7 +157,7 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.proto_json_utils import parse_dict\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..4cd4ffc20 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,7 +39,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _UC_OSS_REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..53975a0af 100644\n--- a/mlflow/store/artifact/artifact_repo.py\n+++ b/mlflow/store/artifact/artifact_repo.py\n@@ -51,22 +51,22 @@ def _truncate_error(err: str, max_length: int = 10_000) -> str:\n return err[:half] + \"\\n\\n*** Error message is too long, truncated ***\\n\\n\" + err[-half:]\n \n \n-def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n- \"\"\"\n- Attempt the try_func with the original credentials (og_creds) if provided, or by generating the\n- credentials using creds_func. If the try_func throws, then try again with new credentials\n- provided by creds_func.\n- \"\"\"\n- try:\n- first_creds = creds_func() if orig_creds is None else orig_creds\n- return try_func(first_creds)\n- except Exception as e:\n- _logger.info(\n- f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n- \" Refreshing credentials and trying again...\"\n- )\n- new_creds = creds_func()\n- return try_func(new_creds)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -78,29 +78,29 @@ class ArtifactRepository:\n \n __metaclass__ = ABCMeta\n \n- def __init__(\n- self, artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None\n- ) -> None:\n- self.artifact_uri = artifact_uri\n- self.tracking_uri = tracking_uri\n- self.registry_uri = registry_uri\n- # Limit the number of threads used for artifact uploads/downloads. Use at most\n- # constants._NUM_MAX_THREADS threads or 2 * the number of CPU cores available on the\n- # system (whichever is smaller)\n- self.thread_pool = self._create_thread_pool()\n-\n- def log_artifact_handler(filename, artifact_path=None, artifact=None):\n- with tempfile.TemporaryDirectory() as tmp_dir:\n- tmp_path = os.path.join(tmp_dir, filename)\n- if artifact is not None:\n- # User should already have installed PIL to log a PIL image\n- from PIL import Image\n-\n- if isinstance(artifact, Image.Image):\n- artifact.save(tmp_path)\n- self.log_artifact(tmp_path, artifact_path)\n-\n- self._async_logging_queue = AsyncArtifactsLoggingQueue(log_artifact_handler)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __repr__(self) -> str:\n return (\n@@ -111,10 +111,10 @@ class ArtifactRepository:\n f\")\"\n )\n \n- def _create_thread_pool(self):\n- return ThreadPoolExecutor(\n- max_workers=self.max_workers, thread_name_prefix=f\"Mlflow{self.__class__.__name__}\"\n- )\n+\n+\n+\n+\n \n def flush_async_logging(self):\n \"\"\"\n@@ -355,11 +355,11 @@ class ArtifactRepository:\n artifact_path: Path of the artifact to delete.\n \"\"\"\n \n- @property\n- def max_workers(self) -> int:\n- \"\"\"Compute the number of workers to use for multi-threading.\"\"\"\n- num_cpus = os.cpu_count() or _NUM_DEFAULT_CPUS\n- return min(num_cpus * _NUM_MAX_THREADS_PER_CPU, _NUM_MAX_THREADS)\n+\n+\n+\n+\n+\n \n def download_trace_data(self) -> dict[str, Any]:\n \"\"\"\n@@ -473,4 +473,4 @@ def verify_artifact_path(artifact_path):\n if artifact_path and path_not_unique(artifact_path):\n raise MlflowException(\n f\"Invalid artifact path: '{artifact_path}'. {bad_path_message(artifact_path)}\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/store/artifact/azure_data_lake_artifact_repo.py b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\nindex 3d34f5971..b4d1dde04 100644\n--- a/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..2b13f2e8f 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,7 +62,7 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/gcs_artifact_repo.py b/mlflow/store/artifact/gcs_artifact_repo.py\nindex c68e7ac82..17febe6bb 100644\n--- a/mlflow/store/artifact/gcs_artifact_repo.py\n+++ b/mlflow/store/artifact/gcs_artifact_repo.py\n@@ -18,11 +18,11 @@ from mlflow.environment_variables import (\n MLFLOW_GCS_UPLOAD_CHUNK_SIZE,\n )\n from mlflow.exceptions import _UnsupportedMultipartUploadException\n-from mlflow.store.artifact.artifact_repo import (\n- ArtifactRepository,\n- MultipartUploadMixin,\n- _retry_with_new_creds,\n-)\n+from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+\n+\n+\n+\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n \n \ndiff --git a/mlflow/store/artifact/optimized_s3_artifact_repo.py b/mlflow/store/artifact/optimized_s3_artifact_repo.py\nindex f40f27abb..76b56ec82 100644\n--- a/mlflow/store/artifact/optimized_s3_artifact_repo.py\n+++ b/mlflow/store/artifact/optimized_s3_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..53db7931d 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -15,10 +15,10 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n ListDirectoryResponse,\n )\n from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..989d8bc69 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,7 +27,7 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..5e871abf6 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,7 +24,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _UC_OSS_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..2a76bb66f 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,7 +37,7 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..da41b2531 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,7 +52,7 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V4_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..4b3652f73 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,7 +120,7 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V3_TRACE_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..f37d92aa6 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,7 +43,7 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/utils/async_logging/async_artifacts_logging_queue.py b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\nindex 08c89366a..87a54662a 100644\n--- a/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n+++ b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n@@ -31,15 +31,15 @@ class AsyncArtifactsLoggingQueue:\n - artifact: The artifact to be logged.\n \"\"\"\n \n- def __init__(\n- self, artifact_logging_func: Callable[[str, str, Union[\"PIL.Image.Image\"]], None]\n- ) -> None:\n- self._queue: Queue[RunArtifact] = Queue()\n- self._lock = threading.RLock()\n- self._artifact_logging_func = artifact_logging_func\n \n- self._stop_data_logging_thread_event = threading.Event()\n- self._is_activated = False\n+\n+\n+\n+\n+\n+\n+\n+\n \n def _at_exit_callback(self) -> None:\n \"\"\"Callback function to be executed when the program is exiting.\n@@ -255,4 +255,4 @@ class AsyncArtifactsLoggingQueue:\n self._set_up_logging_thread()\n atexit.register(self._at_exit_callback)\n \n- self._is_activated = True\n+ self._is_activated = True\n\\ No newline at end of file\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..61480cb97 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -44,89 +44,89 @@ def _mark_int64_fields_for_proto_maps(proto_map, value_field_type):\n return json_dict\n \n \n-def _mark_int64_fields(proto_message):\n- \"\"\"Converts a proto message to JSON, preserving only int64-related fields.\"\"\"\n- json_dict = {}\n- for field, value in proto_message.ListFields():\n- if (\n- # These three conditions check if this field is a protobuf map.\n- # See the official implementation: https://bit.ly/3EMx1rl\n- field.type == FieldDescriptor.TYPE_MESSAGE\n- and field.message_type.has_options\n- and field.message_type.GetOptions().map_entry\n- ):\n- # Deal with proto map fields separately in another function.\n- json_dict[field.name] = _mark_int64_fields_for_proto_maps(\n- value, field.message_type.fields_by_name[\"value\"].type\n- )\n- continue\n \n- if field.type == FieldDescriptor.TYPE_MESSAGE:\n- ftype = partial(_mark_int64_fields)\n- elif field.type in _PROTOBUF_INT64_FIELDS:\n- ftype = int\n- else:\n- # Skip all non-int64 fields.\n- continue\n \n- # Use is_repeated property (preferred) with fallback to deprecated label\n- try:\n- is_repeated = field.is_repeated\n- except AttributeError:\n- is_repeated = field.label == FieldDescriptor.LABEL_REPEATED\n \n- json_dict[field.name] = [ftype(v) for v in value] if is_repeated else ftype(value)\n- return json_dict\n \n \n-def _merge_json_dicts(from_dict, to_dict):\n- \"\"\"Merges the json elements of from_dict into to_dict. Only works for json dicts\n- converted from proto messages\n- \"\"\"\n- for key, value in from_dict.items():\n- if isinstance(key, int) and str(key) in to_dict:\n- # When the key (i.e. the proto field name) is an integer, it must be a proto map field\n- # with integer as the key. For example:\n- # from_dict is {'field_map': {1: '2', 3: '4'}}\n- # to_dict is {'field_map': {'1': '2', '3': '4'}}\n- # So we need to replace the str keys with int keys in to_dict.\n- to_dict[key] = to_dict[str(key)]\n- del to_dict[str(key)]\n-\n- if key not in to_dict:\n- continue\n-\n- if isinstance(value, dict):\n- _merge_json_dicts(from_dict[key], to_dict[key])\n- elif isinstance(value, list):\n- for i, v in enumerate(value):\n- if isinstance(v, dict):\n- _merge_json_dicts(v, to_dict[key][i])\n- else:\n- to_dict[key][i] = v\n- else:\n- to_dict[key] = from_dict[key]\n- return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py\nindex 5d5de9527..aa5b3c40a 100644\n--- a/tests/store/artifact/test_presigned_url_artifact_repo.py\n+++ b/tests/store/artifact/test_presigned_url_artifact_repo.py\n@@ -20,13 +20,13 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n HttpHeader,\n ListDirectoryResponse,\n )\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.presigned_url_artifact_repo import (\n DIRECTORIES_ENDPOINT,\n FILESYSTEM_METHOD_TO_INFO,\n PresignedUrlArtifactRepository,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n \n MODEL_NAME = \"catalog.schema.model\"\n MODEL_VERSION = 1\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..e0f99f4b4 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..540e3ca5f 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,7 +157,7 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.proto_json_utils import parse_dict\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..4cd4ffc20 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,7 +39,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _UC_OSS_REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..53975a0af 100644\n+++ a/mlflow/store/artifact/artifact_repo.py\n--- b/mlflow/store/artifact/artifact_repo.py\n@@ -51,22 +51,22 @@ def _truncate_error(err: str, max_length: int = 10_000) -> str:\n return err[:half] + \"\\n\\n*** Error message is too long, truncated ***\\n\\n\" + err[-half:]\n \n \n+def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n+ \"\"\"\n+ Attempt the try_func with the original credentials (og_creds) if provided, or by generating the\n+ credentials using creds_func. If the try_func throws, then try again with new credentials\n+ provided by creds_func.\n+ \"\"\"\n+ try:\n+ first_creds = creds_func() if orig_creds is None else orig_creds\n+ return try_func(first_creds)\n+ except Exception as e:\n+ _logger.info(\n+ f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n+ \" Refreshing credentials and trying again...\"\n+ )\n+ new_creds = creds_func()\n+ return try_func(new_creds)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -78,29 +78,29 @@ class ArtifactRepository:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(\n+ self, artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None\n+ ) -> None:\n+ self.artifact_uri = artifact_uri\n+ self.tracking_uri = tracking_uri\n+ self.registry_uri = registry_uri\n+ # Limit the number of threads used for artifact uploads/downloads. Use at most\n+ # constants._NUM_MAX_THREADS threads or 2 * the number of CPU cores available on the\n+ # system (whichever is smaller)\n+ self.thread_pool = self._create_thread_pool()\n+\n+ def log_artifact_handler(filename, artifact_path=None, artifact=None):\n+ with tempfile.TemporaryDirectory() as tmp_dir:\n+ tmp_path = os.path.join(tmp_dir, filename)\n+ if artifact is not None:\n+ # User should already have installed PIL to log a PIL image\n+ from PIL import Image\n+\n+ if isinstance(artifact, Image.Image):\n+ artifact.save(tmp_path)\n+ self.log_artifact(tmp_path, artifact_path)\n+\n+ self._async_logging_queue = AsyncArtifactsLoggingQueue(log_artifact_handler)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __repr__(self) -> str:\n return (\n@@ -111,10 +111,10 @@ class ArtifactRepository:\n f\")\"\n )\n \n+ def _create_thread_pool(self):\n+ return ThreadPoolExecutor(\n+ max_workers=self.max_workers, thread_name_prefix=f\"Mlflow{self.__class__.__name__}\"\n+ )\n-\n-\n-\n-\n \n def flush_async_logging(self):\n \"\"\"\n@@ -355,11 +355,11 @@ class ArtifactRepository:\n artifact_path: Path of the artifact to delete.\n \"\"\"\n \n+ @property\n+ def max_workers(self) -> int:\n+ \"\"\"Compute the number of workers to use for multi-threading.\"\"\"\n+ num_cpus = os.cpu_count() or _NUM_DEFAULT_CPUS\n+ return min(num_cpus * _NUM_MAX_THREADS_PER_CPU, _NUM_MAX_THREADS)\n-\n-\n-\n-\n-\n \n def download_trace_data(self) -> dict[str, Any]:\n \"\"\"\n@@ -473,4 +473,4 @@ def verify_artifact_path(artifact_path):\n if artifact_path and path_not_unique(artifact_path):\n raise MlflowException(\n f\"Invalid artifact path: '{artifact_path}'. {bad_path_message(artifact_path)}\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/store/artifact/azure_data_lake_artifact_repo.py b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\nindex 3d34f5971..b4d1dde04 100644\n+++ a/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n--- b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..2b13f2e8f 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,7 +62,7 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/gcs_artifact_repo.py b/mlflow/store/artifact/gcs_artifact_repo.py\nindex c68e7ac82..17febe6bb 100644\n+++ a/mlflow/store/artifact/gcs_artifact_repo.py\n--- b/mlflow/store/artifact/gcs_artifact_repo.py\n@@ -18,11 +18,11 @@ from mlflow.environment_variables import (\n MLFLOW_GCS_UPLOAD_CHUNK_SIZE,\n )\n from mlflow.exceptions import _UnsupportedMultipartUploadException\n+from mlflow.store.artifact.artifact_repo import (\n+ ArtifactRepository,\n+ MultipartUploadMixin,\n+ _retry_with_new_creds,\n+)\n-from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-\n-\n-\n-\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n \n \ndiff --git a/mlflow/store/artifact/optimized_s3_artifact_repo.py b/mlflow/store/artifact/optimized_s3_artifact_repo.py\nindex f40f27abb..76b56ec82 100644\n+++ a/mlflow/store/artifact/optimized_s3_artifact_repo.py\n--- b/mlflow/store/artifact/optimized_s3_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..53db7931d 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -15,10 +15,10 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n ListDirectoryResponse,\n )\n from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..989d8bc69 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,7 +27,7 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..5e871abf6 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,7 +24,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _UC_OSS_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..2a76bb66f 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,7 +37,7 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..da41b2531 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,7 +52,7 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V4_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..4b3652f73 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,7 +120,7 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V3_TRACE_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..f37d92aa6 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,7 +43,7 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/utils/async_logging/async_artifacts_logging_queue.py b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\nindex 08c89366a..87a54662a 100644\n+++ a/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n--- b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n@@ -31,15 +31,15 @@ class AsyncArtifactsLoggingQueue:\n - artifact: The artifact to be logged.\n \"\"\"\n \n+ def __init__(\n+ self, artifact_logging_func: Callable[[str, str, Union[\"PIL.Image.Image\"]], None]\n+ ) -> None:\n+ self._queue: Queue[RunArtifact] = Queue()\n+ self._lock = threading.RLock()\n+ self._artifact_logging_func = artifact_logging_func\n \n+ self._stop_data_logging_thread_event = threading.Event()\n+ self._is_activated = False\n-\n-\n-\n-\n-\n-\n-\n-\n \n def _at_exit_callback(self) -> None:\n \"\"\"Callback function to be executed when the program is exiting.\n@@ -255,4 +255,4 @@ class AsyncArtifactsLoggingQueue:\n self._set_up_logging_thread()\n atexit.register(self._at_exit_callback)\n \n+ self._is_activated = True\n- self._is_activated = True\n\\ No newline at end of file\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..61480cb97 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -44,89 +44,89 @@ def _mark_int64_fields_for_proto_maps(proto_map, value_field_type):\n return json_dict\n \n \n+def _mark_int64_fields(proto_message):\n+ \"\"\"Converts a proto message to JSON, preserving only int64-related fields.\"\"\"\n+ json_dict = {}\n+ for field, value in proto_message.ListFields():\n+ if (\n+ # These three conditions check if this field is a protobuf map.\n+ # See the official implementation: https://bit.ly/3EMx1rl\n+ field.type == FieldDescriptor.TYPE_MESSAGE\n+ and field.message_type.has_options\n+ and field.message_type.GetOptions().map_entry\n+ ):\n+ # Deal with proto map fields separately in another function.\n+ json_dict[field.name] = _mark_int64_fields_for_proto_maps(\n+ value, field.message_type.fields_by_name[\"value\"].type\n+ )\n+ continue\n \n+ if field.type == FieldDescriptor.TYPE_MESSAGE:\n+ ftype = partial(_mark_int64_fields)\n+ elif field.type in _PROTOBUF_INT64_FIELDS:\n+ ftype = int\n+ else:\n+ # Skip all non-int64 fields.\n+ continue\n \n+ # Use is_repeated property (preferred) with fallback to deprecated label\n+ try:\n+ is_repeated = field.is_repeated\n+ except AttributeError:\n+ is_repeated = field.label == FieldDescriptor.LABEL_REPEATED\n \n+ json_dict[field.name] = [ftype(v) for v in value] if is_repeated else ftype(value)\n+ return json_dict\n \n \n+def _merge_json_dicts(from_dict, to_dict):\n+ \"\"\"Merges the json elements of from_dict into to_dict. Only works for json dicts\n+ converted from proto messages\n+ \"\"\"\n+ for key, value in from_dict.items():\n+ if isinstance(key, int) and str(key) in to_dict:\n+ # When the key (i.e. the proto field name) is an integer, it must be a proto map field\n+ # with integer as the key. For example:\n+ # from_dict is {'field_map': {1: '2', 3: '4'}}\n+ # to_dict is {'field_map': {'1': '2', '3': '4'}}\n+ # So we need to replace the str keys with int keys in to_dict.\n+ to_dict[key] = to_dict[str(key)]\n+ del to_dict[str(key)]\n+\n+ if key not in to_dict:\n+ continue\n+\n+ if isinstance(value, dict):\n+ _merge_json_dicts(from_dict[key], to_dict[key])\n+ elif isinstance(value, list):\n+ for i, v in enumerate(value):\n+ if isinstance(v, dict):\n+ _merge_json_dicts(v, to_dict[key][i])\n+ else:\n+ to_dict[key][i] = v\n+ else:\n+ to_dict[key] = from_dict[key]\n+ return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..8dc288245 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -126,7 +126,7 @@ def message_to_json(message):\n json_dict_with_int64_as_numbers = _merge_json_dicts(\n json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=4)\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py\ndeleted file mode 100644\nindex 5d5de9527..000000000\n--- a/tests/store/artifact/test_presigned_url_artifact_repo.py\n+++ /dev/null\n@@ -1,363 +0,0 @@\n-import json\n-import os\n-import random\n-import string\n-from unittest import mock\n-from unittest.mock import ANY\n-\n-import pytest\n-import requests\n-\n-from mlflow.environment_variables import MLFLOW_MULTIPART_DOWNLOAD_CHUNK_SIZE\n-from mlflow.exceptions import RestException\n-from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.protos.databricks_filesystem_service_pb2 import (\n- CreateDownloadUrlRequest,\n- CreateDownloadUrlResponse,\n- CreateUploadUrlRequest,\n- CreateUploadUrlResponse,\n- DirectoryEntry,\n- HttpHeader,\n- ListDirectoryResponse,\n-)\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-from mlflow.store.artifact.presigned_url_artifact_repo import (\n- DIRECTORIES_ENDPOINT,\n- FILESYSTEM_METHOD_TO_INFO,\n- PresignedUrlArtifactRepository,\n-)\n-from mlflow.utils.proto_json_utils import message_to_json\n-\n-MODEL_NAME = \"catalog.schema.model\"\n-MODEL_VERSION = 1\n-MODEL_URI = \"/Models/catalog/schema/model/1\"\n-PRESIGNED_URL_ARTIFACT_REPOSITORY = \"mlflow.store.artifact.presigned_url_artifact_repo\"\n-_DATABRICKS_UC_SCHEME = \"databricks-uc\"\n-\n-\n-@pytest.fixture(autouse=True)\n-def run_around_tests():\n- # mock this call to credentials for all tests in suite\n- with mock.patch(\"mlflow.utils.databricks_utils.get_databricks_host_creds\"):\n- yield\n-\n-\n-def test_artifact_uri():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- assert MODEL_URI == artifact_repo.artifact_uri\n-\n-\n-def mock_list_directory(*args, **kwargs):\n- endpoint = kwargs[\"endpoint\"]\n- json_body = kwargs[\"json_body\"]\n-\n- if endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/dir\" and json_body == json.dumps(\n- {\"page_token\": \"some_token\"}\n- ):\n- return ListDirectoryResponse(\n- contents=[\n- DirectoryEntry(is_directory=False, path=f\"{MODEL_URI}/dir/file2\", file_size=2)\n- ]\n- )\n- elif endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/dir\":\n- return ListDirectoryResponse(\n- contents=[\n- DirectoryEntry(is_directory=False, path=f\"{MODEL_URI}/dir/file1\", file_size=1)\n- ],\n- next_page_token=\"some_token\",\n- )\n- elif endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/\":\n- return ListDirectoryResponse(\n- contents=[DirectoryEntry(is_directory=True, path=f\"{MODEL_URI}/dir\")],\n- )\n- else:\n- raise ValueError(f\"Unexpected endpoint: {endpoint}\")\n-\n-\n-def test_list_artifact_pagination():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_list_directory\n- ) as mock_list:\n- resp = artifact_repo.list_artifacts()\n- assert len(resp) == 1\n- assert resp[0].path == \"dir\"\n- assert resp[0].is_dir is True\n- assert resp[0].file_size is None\n-\n- resp = artifact_repo.list_artifacts(\"dir\")\n- assert len(resp) == 2\n- assert {r.path for r in resp} == {\"dir/file1\", \"dir/file2\"}\n- assert {r.is_dir for r in resp} == {False}\n- assert {r.file_size for r in resp} == {1, 2}\n-\n- assert mock_list.call_count == 3\n-\n-\n-def test_list_artifacts_failure():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- exc_code = \"NOT_FOUND\"\n- exc_message = \"The directory being accessed is not found.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc):\n- empty_infos = artifact_repo.list_artifacts(remote_file_path)\n- assert len(empty_infos) == 0\n-\n-\n-def _make_presigned_url(remote_path):\n- return f\"presigned_url/{remote_path}\"\n-\n-\n-def _make_headers(remote_path):\n- return {f\"path_header-{remote_path}\": f\"remote-path={remote_path}\"}\n-\n-\n-def mock_create_download_url(*args, **kwargs):\n- remote_path = json.loads(kwargs[\"json_body\"])[\"path\"]\n- return CreateDownloadUrlResponse(\n- url=_make_presigned_url(remote_path),\n- headers=[\n- HttpHeader(name=header, value=val) for header, val in _make_headers(remote_path).items()\n- ],\n- )\n-\n-\n-def test_get_read_credentials():\n- remote_file_paths = [\"file\", \"dir/file1\", \"dir/file2\"]\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- download_url_endpoint = \"/api/2.0/fs/create-download-url\"\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_create_download_url\n- ) as mock_call_endpoint:\n- creds = artifact_repo._get_read_credential_infos(remote_file_paths)\n- assert mock_call_endpoint.call_count == 3\n-\n- for remote_file_path in remote_file_paths:\n- mock_call_endpoint.assert_any_call(\n- host_creds=ANY,\n- endpoint=f\"{download_url_endpoint}\",\n- method=\"POST\",\n- json_body=message_to_json(\n- CreateDownloadUrlRequest(path=f\"{MODEL_URI}/{remote_file_path}\")\n- ),\n- response_proto=ANY,\n- )\n-\n- assert {_make_presigned_url(f\"{MODEL_URI}/{path}\") for path in remote_file_paths} == {\n- cred.signed_uri for cred in creds\n- }\n- expected_headers = {}\n- for path in remote_file_paths:\n- expected_headers.update(_make_headers(f\"{MODEL_URI}/{path}\"))\n- actual_headers = {}\n- for cred in creds:\n- actual_headers.update({header.name: header.value for header in cred.headers})\n- assert expected_headers == actual_headers\n-\n-\n-def mock_create_upload_url(*args, **kwargs):\n- remote_path = json.loads(kwargs[\"json_body\"])[\"path\"]\n- return CreateUploadUrlResponse(\n- url=_make_presigned_url(remote_path),\n- headers=[\n- HttpHeader(name=header, value=val) for header, val in _make_headers(remote_path).items()\n- ],\n- )\n-\n-\n-def test_get_write_credentials():\n- remote_file_paths = [\"file\", \"dir/file1\", \"dir/file2\"]\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- upload_url_endpoint = \"/api/2.0/fs/create-upload-url\"\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_create_upload_url\n- ) as mock_call_endpoint:\n- creds = artifact_repo._get_write_credential_infos(remote_file_paths)\n- assert mock_call_endpoint.call_count == 3\n- for remote_file_path in remote_file_paths:\n- mock_call_endpoint.assert_any_call(\n- host_creds=ANY,\n- endpoint=f\"{upload_url_endpoint}\",\n- method=\"POST\",\n- json_body=message_to_json(\n- CreateUploadUrlRequest(path=f\"{MODEL_URI}/{remote_file_path}\")\n- ),\n- response_proto=ANY,\n- )\n-\n- expected_headers = {}\n- for path in remote_file_paths:\n- expected_headers.update(_make_headers(f\"{MODEL_URI}/{path}\"))\n- actual_headers = {}\n- for cred in creds:\n- actual_headers.update({header.name: header.value for header in cred.headers})\n- assert expected_headers == actual_headers\n-\n-\n-def test_download_from_cloud():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._get_download_presigned_url_and_headers\",\n- return_value=CreateDownloadUrlResponse(\n- url=_make_presigned_url(remote_file_path),\n- headers=[\n- HttpHeader(name=k, value=v) for k, v in _make_headers(remote_file_path).items()\n- ],\n- ),\n- ) as mock_request,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.download_file_using_http_uri\"\n- ) as mock_download,\n- ):\n- local_file = \"local_file\"\n- artifact_repo._download_from_cloud(remote_file_path, local_file)\n-\n- mock_request.assert_called_once_with(remote_file_path)\n- mock_download.assert_called_once_with(\n- http_uri=_make_presigned_url(remote_file_path),\n- download_path=local_file,\n- chunk_size=MLFLOW_MULTIPART_DOWNLOAD_CHUNK_SIZE.get(),\n- headers=_make_headers(remote_file_path),\n- )\n-\n-\n-def test_download_from_cloud_fail():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- endpoint, _ = FILESYSTEM_METHOD_TO_INFO[CreateDownloadUrlRequest]\n- exc_code = \"ENDPOINT_NOT_FOUND\"\n- exc_message = f\"Endpoint not found for {endpoint.lstrip('api')}.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with (\n- mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc),\n- pytest.raises(RestException) as exc_info, # noqa: PT011\n- ):\n- artifact_repo._download_from_cloud(remote_file_path, \"local_file\")\n-\n- assert exc_info.value.error_code == exc_code\n- assert str(exc_info.value) == f\"{exc_code}: {exc_message}\"\n-\n-\n-def test_log_artifact():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- local_file = \"local_file\"\n- artifact_path = \"remote/file/location\"\n- total_remote_path = f\"{artifact_path}/{os.path.basename(local_file)}\"\n- creds = ArtifactCredentialInfo(\n- signed_uri=_make_presigned_url(total_remote_path),\n- headers=[\n- ArtifactCredentialInfo.HttpHeader(name=k, value=v)\n- for k, v in _make_headers(total_remote_path).items()\n- ],\n- )\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._get_write_credential_infos\",\n- return_value=[creds],\n- ) as mock_request,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._upload_to_cloud\",\n- return_value=None,\n- ) as mock_upload,\n- ):\n- artifact_repo.log_artifact(local_file, artifact_path)\n- mock_request.assert_called_once_with(remote_file_paths=[total_remote_path])\n- mock_upload.assert_called_once_with(\n- cloud_credential_info=creds,\n- src_file_path=local_file,\n- artifact_file_path=total_remote_path,\n- )\n-\n-\n-def test_upload_to_cloud(tmp_path):\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- local_file = os.path.join(tmp_path, \"file.txt\")\n- content = \"content\"\n- with open(local_file, \"w\") as f:\n- f.write(content)\n- remote_file_path = \"some/remote/file/path\"\n- resp = mock.create_autospec(requests.Response, return_value=None)\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.cloud_storage_http_request\", return_value=resp\n- ) as mock_cloud,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.augmented_raise_for_status\"\n- ) as mock_status,\n- ):\n- cred_info = ArtifactCredentialInfo(\n- signed_uri=_make_presigned_url(remote_file_path),\n- headers=[\n- ArtifactCredentialInfo.HttpHeader(name=k, value=v)\n- for k, v in _make_headers(remote_file_path).items()\n- ],\n- )\n- artifact_repo._upload_to_cloud(cred_info, local_file, \"some/irrelevant/path\")\n- mock_cloud.assert_called_once_with(\n- \"put\",\n- _make_presigned_url(remote_file_path),\n- data=bytearray(content, \"utf-8\"),\n- headers=_make_headers(remote_file_path),\n- )\n- mock_status.assert_called_once_with(resp.__enter__())\n-\n-\n-def test_upload_to_cloud_fail():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- endpoint, _ = FILESYSTEM_METHOD_TO_INFO[CreateUploadUrlRequest]\n- exc_code = \"ENDPOINT_NOT_FOUND\"\n- exc_message = f\"Endpoint not found for {endpoint.lstrip('api')}.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with (\n- mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc),\n- pytest.raises(RestException) as exc_info, # noqa: PT011\n- ):\n- artifact_repo._download_from_cloud(remote_file_path, \"local_file\")\n-\n- assert exc_info.value.error_code == exc_code\n- assert str(exc_info.value) == f\"{exc_code}: {exc_message}\"\n-\n-\n-@pytest.mark.parametrize(\"throw\", [True, False])\n-@pytest.mark.parametrize(\"use_og_creds\", [True, False])\n-@pytest.mark.parametrize(\"status_code\", [401, 403])\n-def test_retry_refresh_creds_no_creds(throw, use_og_creds, status_code):\n- credentials = \"og_creds\"\n- called = False\n-\n- def creds_func():\n- nonlocal credentials\n- credentials = \"\".join(random.choices(string.ascii_lowercase, k=10))\n- return credentials\n-\n- def try_func(creds):\n- nonlocal called, credentials\n- assert creds == credentials\n- resp = requests.Response()\n- resp.status_code = status_code\n- if throw and not called:\n- called = True\n- raise requests.HTTPError(response=resp)\n-\n- mock_creds = mock.Mock(side_effect=creds_func)\n- mock_func = mock.Mock(side_effect=try_func)\n- if use_og_creds:\n- _retry_with_new_creds(try_func=mock_func, creds_func=mock_creds, orig_creds=credentials)\n- else:\n- _retry_with_new_creds(try_func=mock_func, creds_func=mock_creds)\n-\n- if throw:\n- assert mock_func.call_count == 2\n- assert mock_creds.call_count == 1 if use_og_creds else 2\n- else:\n- assert mock_func.call_count == 1\n- assert mock_creds.call_count == 0 if use_og_creds else 1\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S16\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S18\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S19\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n }\n ]\n}"}}},{"rowIdx":94,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_presigned_url_artifact_repo.8d57288e.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S19\",\n \"text\": \"Uses Google's MessageToJson API internally but applies additional processing to handle int64 fields correctly.\\n\\nBy default, Google's MessageToJson converts int64 proto fields to JSON strings for compatibility reasons, but this function converts them back to numbers\",\n \"category\": \"implementation_constraint\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/utils/proto_json_utils.py\",\n \"lines\": \"113-127\",\n \"function\": \"message_to_json\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..e0f99f4b4 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..540e3ca5f 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,7 +157,7 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.proto_json_utils import parse_dict\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..4cd4ffc20 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,7 +39,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _UC_OSS_REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..53975a0af 100644\n--- a/mlflow/store/artifact/artifact_repo.py\n+++ b/mlflow/store/artifact/artifact_repo.py\n@@ -51,22 +51,22 @@ def _truncate_error(err: str, max_length: int = 10_000) -> str:\n return err[:half] + \"\\n\\n*** Error message is too long, truncated ***\\n\\n\" + err[-half:]\n \n \n-def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n- \"\"\"\n- Attempt the try_func with the original credentials (og_creds) if provided, or by generating the\n- credentials using creds_func. If the try_func throws, then try again with new credentials\n- provided by creds_func.\n- \"\"\"\n- try:\n- first_creds = creds_func() if orig_creds is None else orig_creds\n- return try_func(first_creds)\n- except Exception as e:\n- _logger.info(\n- f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n- \" Refreshing credentials and trying again...\"\n- )\n- new_creds = creds_func()\n- return try_func(new_creds)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -78,29 +78,29 @@ class ArtifactRepository:\n \n __metaclass__ = ABCMeta\n \n- def __init__(\n- self, artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None\n- ) -> None:\n- self.artifact_uri = artifact_uri\n- self.tracking_uri = tracking_uri\n- self.registry_uri = registry_uri\n- # Limit the number of threads used for artifact uploads/downloads. Use at most\n- # constants._NUM_MAX_THREADS threads or 2 * the number of CPU cores available on the\n- # system (whichever is smaller)\n- self.thread_pool = self._create_thread_pool()\n-\n- def log_artifact_handler(filename, artifact_path=None, artifact=None):\n- with tempfile.TemporaryDirectory() as tmp_dir:\n- tmp_path = os.path.join(tmp_dir, filename)\n- if artifact is not None:\n- # User should already have installed PIL to log a PIL image\n- from PIL import Image\n-\n- if isinstance(artifact, Image.Image):\n- artifact.save(tmp_path)\n- self.log_artifact(tmp_path, artifact_path)\n-\n- self._async_logging_queue = AsyncArtifactsLoggingQueue(log_artifact_handler)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __repr__(self) -> str:\n return (\n@@ -111,10 +111,10 @@ class ArtifactRepository:\n f\")\"\n )\n \n- def _create_thread_pool(self):\n- return ThreadPoolExecutor(\n- max_workers=self.max_workers, thread_name_prefix=f\"Mlflow{self.__class__.__name__}\"\n- )\n+\n+\n+\n+\n \n def flush_async_logging(self):\n \"\"\"\n@@ -355,11 +355,11 @@ class ArtifactRepository:\n artifact_path: Path of the artifact to delete.\n \"\"\"\n \n- @property\n- def max_workers(self) -> int:\n- \"\"\"Compute the number of workers to use for multi-threading.\"\"\"\n- num_cpus = os.cpu_count() or _NUM_DEFAULT_CPUS\n- return min(num_cpus * _NUM_MAX_THREADS_PER_CPU, _NUM_MAX_THREADS)\n+\n+\n+\n+\n+\n \n def download_trace_data(self) -> dict[str, Any]:\n \"\"\"\n@@ -473,4 +473,4 @@ def verify_artifact_path(artifact_path):\n if artifact_path and path_not_unique(artifact_path):\n raise MlflowException(\n f\"Invalid artifact path: '{artifact_path}'. {bad_path_message(artifact_path)}\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/store/artifact/azure_data_lake_artifact_repo.py b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\nindex 3d34f5971..b4d1dde04 100644\n--- a/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..2b13f2e8f 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,7 +62,7 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/gcs_artifact_repo.py b/mlflow/store/artifact/gcs_artifact_repo.py\nindex c68e7ac82..17febe6bb 100644\n--- a/mlflow/store/artifact/gcs_artifact_repo.py\n+++ b/mlflow/store/artifact/gcs_artifact_repo.py\n@@ -18,11 +18,11 @@ from mlflow.environment_variables import (\n MLFLOW_GCS_UPLOAD_CHUNK_SIZE,\n )\n from mlflow.exceptions import _UnsupportedMultipartUploadException\n-from mlflow.store.artifact.artifact_repo import (\n- ArtifactRepository,\n- MultipartUploadMixin,\n- _retry_with_new_creds,\n-)\n+from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+\n+\n+\n+\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n \n \ndiff --git a/mlflow/store/artifact/optimized_s3_artifact_repo.py b/mlflow/store/artifact/optimized_s3_artifact_repo.py\nindex f40f27abb..76b56ec82 100644\n--- a/mlflow/store/artifact/optimized_s3_artifact_repo.py\n+++ b/mlflow/store/artifact/optimized_s3_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..53db7931d 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -15,10 +15,10 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n ListDirectoryResponse,\n )\n from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..989d8bc69 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,7 +27,7 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..5e871abf6 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,7 +24,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _UC_OSS_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..2a76bb66f 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,7 +37,7 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..da41b2531 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,7 +52,7 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V4_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..4b3652f73 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,7 +120,7 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V3_TRACE_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..f37d92aa6 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,7 +43,7 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/utils/async_logging/async_artifacts_logging_queue.py b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\nindex 08c89366a..87a54662a 100644\n--- a/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n+++ b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n@@ -31,15 +31,15 @@ class AsyncArtifactsLoggingQueue:\n - artifact: The artifact to be logged.\n \"\"\"\n \n- def __init__(\n- self, artifact_logging_func: Callable[[str, str, Union[\"PIL.Image.Image\"]], None]\n- ) -> None:\n- self._queue: Queue[RunArtifact] = Queue()\n- self._lock = threading.RLock()\n- self._artifact_logging_func = artifact_logging_func\n \n- self._stop_data_logging_thread_event = threading.Event()\n- self._is_activated = False\n+\n+\n+\n+\n+\n+\n+\n+\n \n def _at_exit_callback(self) -> None:\n \"\"\"Callback function to be executed when the program is exiting.\n@@ -255,4 +255,4 @@ class AsyncArtifactsLoggingQueue:\n self._set_up_logging_thread()\n atexit.register(self._at_exit_callback)\n \n- self._is_activated = True\n+ self._is_activated = True\n\\ No newline at end of file\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..61480cb97 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -44,89 +44,89 @@ def _mark_int64_fields_for_proto_maps(proto_map, value_field_type):\n return json_dict\n \n \n-def _mark_int64_fields(proto_message):\n- \"\"\"Converts a proto message to JSON, preserving only int64-related fields.\"\"\"\n- json_dict = {}\n- for field, value in proto_message.ListFields():\n- if (\n- # These three conditions check if this field is a protobuf map.\n- # See the official implementation: https://bit.ly/3EMx1rl\n- field.type == FieldDescriptor.TYPE_MESSAGE\n- and field.message_type.has_options\n- and field.message_type.GetOptions().map_entry\n- ):\n- # Deal with proto map fields separately in another function.\n- json_dict[field.name] = _mark_int64_fields_for_proto_maps(\n- value, field.message_type.fields_by_name[\"value\"].type\n- )\n- continue\n \n- if field.type == FieldDescriptor.TYPE_MESSAGE:\n- ftype = partial(_mark_int64_fields)\n- elif field.type in _PROTOBUF_INT64_FIELDS:\n- ftype = int\n- else:\n- # Skip all non-int64 fields.\n- continue\n \n- # Use is_repeated property (preferred) with fallback to deprecated label\n- try:\n- is_repeated = field.is_repeated\n- except AttributeError:\n- is_repeated = field.label == FieldDescriptor.LABEL_REPEATED\n \n- json_dict[field.name] = [ftype(v) for v in value] if is_repeated else ftype(value)\n- return json_dict\n \n \n-def _merge_json_dicts(from_dict, to_dict):\n- \"\"\"Merges the json elements of from_dict into to_dict. Only works for json dicts\n- converted from proto messages\n- \"\"\"\n- for key, value in from_dict.items():\n- if isinstance(key, int) and str(key) in to_dict:\n- # When the key (i.e. the proto field name) is an integer, it must be a proto map field\n- # with integer as the key. For example:\n- # from_dict is {'field_map': {1: '2', 3: '4'}}\n- # to_dict is {'field_map': {'1': '2', '3': '4'}}\n- # So we need to replace the str keys with int keys in to_dict.\n- to_dict[key] = to_dict[str(key)]\n- del to_dict[str(key)]\n-\n- if key not in to_dict:\n- continue\n-\n- if isinstance(value, dict):\n- _merge_json_dicts(from_dict[key], to_dict[key])\n- elif isinstance(value, list):\n- for i, v in enumerate(value):\n- if isinstance(v, dict):\n- _merge_json_dicts(v, to_dict[key][i])\n- else:\n- to_dict[key][i] = v\n- else:\n- to_dict[key] = from_dict[key]\n- return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py\nindex 5d5de9527..aa5b3c40a 100644\n--- a/tests/store/artifact/test_presigned_url_artifact_repo.py\n+++ b/tests/store/artifact/test_presigned_url_artifact_repo.py\n@@ -20,13 +20,13 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n HttpHeader,\n ListDirectoryResponse,\n )\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.presigned_url_artifact_repo import (\n DIRECTORIES_ENDPOINT,\n FILESYSTEM_METHOD_TO_INFO,\n PresignedUrlArtifactRepository,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n \n MODEL_NAME = \"catalog.schema.model\"\n MODEL_VERSION = 1\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..e0f99f4b4 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..540e3ca5f 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,7 +157,7 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.proto_json_utils import parse_dict\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..4cd4ffc20 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,7 +39,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _UC_OSS_REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..53975a0af 100644\n+++ a/mlflow/store/artifact/artifact_repo.py\n--- b/mlflow/store/artifact/artifact_repo.py\n@@ -51,22 +51,22 @@ def _truncate_error(err: str, max_length: int = 10_000) -> str:\n return err[:half] + \"\\n\\n*** Error message is too long, truncated ***\\n\\n\" + err[-half:]\n \n \n+def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n+ \"\"\"\n+ Attempt the try_func with the original credentials (og_creds) if provided, or by generating the\n+ credentials using creds_func. If the try_func throws, then try again with new credentials\n+ provided by creds_func.\n+ \"\"\"\n+ try:\n+ first_creds = creds_func() if orig_creds is None else orig_creds\n+ return try_func(first_creds)\n+ except Exception as e:\n+ _logger.info(\n+ f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n+ \" Refreshing credentials and trying again...\"\n+ )\n+ new_creds = creds_func()\n+ return try_func(new_creds)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -78,29 +78,29 @@ class ArtifactRepository:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(\n+ self, artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None\n+ ) -> None:\n+ self.artifact_uri = artifact_uri\n+ self.tracking_uri = tracking_uri\n+ self.registry_uri = registry_uri\n+ # Limit the number of threads used for artifact uploads/downloads. Use at most\n+ # constants._NUM_MAX_THREADS threads or 2 * the number of CPU cores available on the\n+ # system (whichever is smaller)\n+ self.thread_pool = self._create_thread_pool()\n+\n+ def log_artifact_handler(filename, artifact_path=None, artifact=None):\n+ with tempfile.TemporaryDirectory() as tmp_dir:\n+ tmp_path = os.path.join(tmp_dir, filename)\n+ if artifact is not None:\n+ # User should already have installed PIL to log a PIL image\n+ from PIL import Image\n+\n+ if isinstance(artifact, Image.Image):\n+ artifact.save(tmp_path)\n+ self.log_artifact(tmp_path, artifact_path)\n+\n+ self._async_logging_queue = AsyncArtifactsLoggingQueue(log_artifact_handler)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __repr__(self) -> str:\n return (\n@@ -111,10 +111,10 @@ class ArtifactRepository:\n f\")\"\n )\n \n+ def _create_thread_pool(self):\n+ return ThreadPoolExecutor(\n+ max_workers=self.max_workers, thread_name_prefix=f\"Mlflow{self.__class__.__name__}\"\n+ )\n-\n-\n-\n-\n \n def flush_async_logging(self):\n \"\"\"\n@@ -355,11 +355,11 @@ class ArtifactRepository:\n artifact_path: Path of the artifact to delete.\n \"\"\"\n \n+ @property\n+ def max_workers(self) -> int:\n+ \"\"\"Compute the number of workers to use for multi-threading.\"\"\"\n+ num_cpus = os.cpu_count() or _NUM_DEFAULT_CPUS\n+ return min(num_cpus * _NUM_MAX_THREADS_PER_CPU, _NUM_MAX_THREADS)\n-\n-\n-\n-\n-\n \n def download_trace_data(self) -> dict[str, Any]:\n \"\"\"\n@@ -473,4 +473,4 @@ def verify_artifact_path(artifact_path):\n if artifact_path and path_not_unique(artifact_path):\n raise MlflowException(\n f\"Invalid artifact path: '{artifact_path}'. {bad_path_message(artifact_path)}\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/store/artifact/azure_data_lake_artifact_repo.py b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\nindex 3d34f5971..b4d1dde04 100644\n+++ a/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n--- b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..2b13f2e8f 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,7 +62,7 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/gcs_artifact_repo.py b/mlflow/store/artifact/gcs_artifact_repo.py\nindex c68e7ac82..17febe6bb 100644\n+++ a/mlflow/store/artifact/gcs_artifact_repo.py\n--- b/mlflow/store/artifact/gcs_artifact_repo.py\n@@ -18,11 +18,11 @@ from mlflow.environment_variables import (\n MLFLOW_GCS_UPLOAD_CHUNK_SIZE,\n )\n from mlflow.exceptions import _UnsupportedMultipartUploadException\n+from mlflow.store.artifact.artifact_repo import (\n+ ArtifactRepository,\n+ MultipartUploadMixin,\n+ _retry_with_new_creds,\n+)\n-from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-\n-\n-\n-\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n \n \ndiff --git a/mlflow/store/artifact/optimized_s3_artifact_repo.py b/mlflow/store/artifact/optimized_s3_artifact_repo.py\nindex f40f27abb..76b56ec82 100644\n+++ a/mlflow/store/artifact/optimized_s3_artifact_repo.py\n--- b/mlflow/store/artifact/optimized_s3_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..53db7931d 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -15,10 +15,10 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n ListDirectoryResponse,\n )\n from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..989d8bc69 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,7 +27,7 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..5e871abf6 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,7 +24,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _UC_OSS_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..2a76bb66f 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,7 +37,7 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..da41b2531 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,7 +52,7 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V4_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..4b3652f73 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,7 +120,7 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V3_TRACE_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..f37d92aa6 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,7 +43,7 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/utils/async_logging/async_artifacts_logging_queue.py b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\nindex 08c89366a..87a54662a 100644\n+++ a/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n--- b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n@@ -31,15 +31,15 @@ class AsyncArtifactsLoggingQueue:\n - artifact: The artifact to be logged.\n \"\"\"\n \n+ def __init__(\n+ self, artifact_logging_func: Callable[[str, str, Union[\"PIL.Image.Image\"]], None]\n+ ) -> None:\n+ self._queue: Queue[RunArtifact] = Queue()\n+ self._lock = threading.RLock()\n+ self._artifact_logging_func = artifact_logging_func\n \n+ self._stop_data_logging_thread_event = threading.Event()\n+ self._is_activated = False\n-\n-\n-\n-\n-\n-\n-\n-\n \n def _at_exit_callback(self) -> None:\n \"\"\"Callback function to be executed when the program is exiting.\n@@ -255,4 +255,4 @@ class AsyncArtifactsLoggingQueue:\n self._set_up_logging_thread()\n atexit.register(self._at_exit_callback)\n \n+ self._is_activated = True\n- self._is_activated = True\n\\ No newline at end of file\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..61480cb97 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -44,89 +44,89 @@ def _mark_int64_fields_for_proto_maps(proto_map, value_field_type):\n return json_dict\n \n \n+def _mark_int64_fields(proto_message):\n+ \"\"\"Converts a proto message to JSON, preserving only int64-related fields.\"\"\"\n+ json_dict = {}\n+ for field, value in proto_message.ListFields():\n+ if (\n+ # These three conditions check if this field is a protobuf map.\n+ # See the official implementation: https://bit.ly/3EMx1rl\n+ field.type == FieldDescriptor.TYPE_MESSAGE\n+ and field.message_type.has_options\n+ and field.message_type.GetOptions().map_entry\n+ ):\n+ # Deal with proto map fields separately in another function.\n+ json_dict[field.name] = _mark_int64_fields_for_proto_maps(\n+ value, field.message_type.fields_by_name[\"value\"].type\n+ )\n+ continue\n \n+ if field.type == FieldDescriptor.TYPE_MESSAGE:\n+ ftype = partial(_mark_int64_fields)\n+ elif field.type in _PROTOBUF_INT64_FIELDS:\n+ ftype = int\n+ else:\n+ # Skip all non-int64 fields.\n+ continue\n \n+ # Use is_repeated property (preferred) with fallback to deprecated label\n+ try:\n+ is_repeated = field.is_repeated\n+ except AttributeError:\n+ is_repeated = field.label == FieldDescriptor.LABEL_REPEATED\n \n+ json_dict[field.name] = [ftype(v) for v in value] if is_repeated else ftype(value)\n+ return json_dict\n \n \n+def _merge_json_dicts(from_dict, to_dict):\n+ \"\"\"Merges the json elements of from_dict into to_dict. Only works for json dicts\n+ converted from proto messages\n+ \"\"\"\n+ for key, value in from_dict.items():\n+ if isinstance(key, int) and str(key) in to_dict:\n+ # When the key (i.e. the proto field name) is an integer, it must be a proto map field\n+ # with integer as the key. For example:\n+ # from_dict is {'field_map': {1: '2', 3: '4'}}\n+ # to_dict is {'field_map': {'1': '2', '3': '4'}}\n+ # So we need to replace the str keys with int keys in to_dict.\n+ to_dict[key] = to_dict[str(key)]\n+ del to_dict[str(key)]\n+\n+ if key not in to_dict:\n+ continue\n+\n+ if isinstance(value, dict):\n+ _merge_json_dicts(from_dict[key], to_dict[key])\n+ elif isinstance(value, list):\n+ for i, v in enumerate(value):\n+ if isinstance(v, dict):\n+ _merge_json_dicts(v, to_dict[key][i])\n+ else:\n+ to_dict[key][i] = v\n+ else:\n+ to_dict[key] = from_dict[key]\n+ return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..e94358bdc 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -114,8 +114,9 @@ def message_to_json(message):\n \n # Google's MessageToJson API converts int64 proto fields to JSON strings.\n # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n+ from google.protobuf.json_format import MessageToDict\n+ json_dict_with_int64_as_str = MessageToDict(\n+ message, preserving_proto_field_name=True\n )\n # We convert this proto message into a JSON dict where only int64 proto fields\n # are preserved, and they are treated as JSON numbers, not strings.\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py\ndeleted file mode 100644\nindex 5d5de9527..000000000\n--- a/tests/store/artifact/test_presigned_url_artifact_repo.py\n+++ /dev/null\n@@ -1,363 +0,0 @@\n-import json\n-import os\n-import random\n-import string\n-from unittest import mock\n-from unittest.mock import ANY\n-\n-import pytest\n-import requests\n-\n-from mlflow.environment_variables import MLFLOW_MULTIPART_DOWNLOAD_CHUNK_SIZE\n-from mlflow.exceptions import RestException\n-from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.protos.databricks_filesystem_service_pb2 import (\n- CreateDownloadUrlRequest,\n- CreateDownloadUrlResponse,\n- CreateUploadUrlRequest,\n- CreateUploadUrlResponse,\n- DirectoryEntry,\n- HttpHeader,\n- ListDirectoryResponse,\n-)\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-from mlflow.store.artifact.presigned_url_artifact_repo import (\n- DIRECTORIES_ENDPOINT,\n- FILESYSTEM_METHOD_TO_INFO,\n- PresignedUrlArtifactRepository,\n-)\n-from mlflow.utils.proto_json_utils import message_to_json\n-\n-MODEL_NAME = \"catalog.schema.model\"\n-MODEL_VERSION = 1\n-MODEL_URI = \"/Models/catalog/schema/model/1\"\n-PRESIGNED_URL_ARTIFACT_REPOSITORY = \"mlflow.store.artifact.presigned_url_artifact_repo\"\n-_DATABRICKS_UC_SCHEME = \"databricks-uc\"\n-\n-\n-@pytest.fixture(autouse=True)\n-def run_around_tests():\n- # mock this call to credentials for all tests in suite\n- with mock.patch(\"mlflow.utils.databricks_utils.get_databricks_host_creds\"):\n- yield\n-\n-\n-def test_artifact_uri():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- assert MODEL_URI == artifact_repo.artifact_uri\n-\n-\n-def mock_list_directory(*args, **kwargs):\n- endpoint = kwargs[\"endpoint\"]\n- json_body = kwargs[\"json_body\"]\n-\n- if endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/dir\" and json_body == json.dumps(\n- {\"page_token\": \"some_token\"}\n- ):\n- return ListDirectoryResponse(\n- contents=[\n- DirectoryEntry(is_directory=False, path=f\"{MODEL_URI}/dir/file2\", file_size=2)\n- ]\n- )\n- elif endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/dir\":\n- return ListDirectoryResponse(\n- contents=[\n- DirectoryEntry(is_directory=False, path=f\"{MODEL_URI}/dir/file1\", file_size=1)\n- ],\n- next_page_token=\"some_token\",\n- )\n- elif endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/\":\n- return ListDirectoryResponse(\n- contents=[DirectoryEntry(is_directory=True, path=f\"{MODEL_URI}/dir\")],\n- )\n- else:\n- raise ValueError(f\"Unexpected endpoint: {endpoint}\")\n-\n-\n-def test_list_artifact_pagination():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_list_directory\n- ) as mock_list:\n- resp = artifact_repo.list_artifacts()\n- assert len(resp) == 1\n- assert resp[0].path == \"dir\"\n- assert resp[0].is_dir is True\n- assert resp[0].file_size is None\n-\n- resp = artifact_repo.list_artifacts(\"dir\")\n- assert len(resp) == 2\n- assert {r.path for r in resp} == {\"dir/file1\", \"dir/file2\"}\n- assert {r.is_dir for r in resp} == {False}\n- assert {r.file_size for r in resp} == {1, 2}\n-\n- assert mock_list.call_count == 3\n-\n-\n-def test_list_artifacts_failure():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- exc_code = \"NOT_FOUND\"\n- exc_message = \"The directory being accessed is not found.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc):\n- empty_infos = artifact_repo.list_artifacts(remote_file_path)\n- assert len(empty_infos) == 0\n-\n-\n-def _make_presigned_url(remote_path):\n- return f\"presigned_url/{remote_path}\"\n-\n-\n-def _make_headers(remote_path):\n- return {f\"path_header-{remote_path}\": f\"remote-path={remote_path}\"}\n-\n-\n-def mock_create_download_url(*args, **kwargs):\n- remote_path = json.loads(kwargs[\"json_body\"])[\"path\"]\n- return CreateDownloadUrlResponse(\n- url=_make_presigned_url(remote_path),\n- headers=[\n- HttpHeader(name=header, value=val) for header, val in _make_headers(remote_path).items()\n- ],\n- )\n-\n-\n-def test_get_read_credentials():\n- remote_file_paths = [\"file\", \"dir/file1\", \"dir/file2\"]\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- download_url_endpoint = \"/api/2.0/fs/create-download-url\"\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_create_download_url\n- ) as mock_call_endpoint:\n- creds = artifact_repo._get_read_credential_infos(remote_file_paths)\n- assert mock_call_endpoint.call_count == 3\n-\n- for remote_file_path in remote_file_paths:\n- mock_call_endpoint.assert_any_call(\n- host_creds=ANY,\n- endpoint=f\"{download_url_endpoint}\",\n- method=\"POST\",\n- json_body=message_to_json(\n- CreateDownloadUrlRequest(path=f\"{MODEL_URI}/{remote_file_path}\")\n- ),\n- response_proto=ANY,\n- )\n-\n- assert {_make_presigned_url(f\"{MODEL_URI}/{path}\") for path in remote_file_paths} == {\n- cred.signed_uri for cred in creds\n- }\n- expected_headers = {}\n- for path in remote_file_paths:\n- expected_headers.update(_make_headers(f\"{MODEL_URI}/{path}\"))\n- actual_headers = {}\n- for cred in creds:\n- actual_headers.update({header.name: header.value for header in cred.headers})\n- assert expected_headers == actual_headers\n-\n-\n-def mock_create_upload_url(*args, **kwargs):\n- remote_path = json.loads(kwargs[\"json_body\"])[\"path\"]\n- return CreateUploadUrlResponse(\n- url=_make_presigned_url(remote_path),\n- headers=[\n- HttpHeader(name=header, value=val) for header, val in _make_headers(remote_path).items()\n- ],\n- )\n-\n-\n-def test_get_write_credentials():\n- remote_file_paths = [\"file\", \"dir/file1\", \"dir/file2\"]\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- upload_url_endpoint = \"/api/2.0/fs/create-upload-url\"\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_create_upload_url\n- ) as mock_call_endpoint:\n- creds = artifact_repo._get_write_credential_infos(remote_file_paths)\n- assert mock_call_endpoint.call_count == 3\n- for remote_file_path in remote_file_paths:\n- mock_call_endpoint.assert_any_call(\n- host_creds=ANY,\n- endpoint=f\"{upload_url_endpoint}\",\n- method=\"POST\",\n- json_body=message_to_json(\n- CreateUploadUrlRequest(path=f\"{MODEL_URI}/{remote_file_path}\")\n- ),\n- response_proto=ANY,\n- )\n-\n- expected_headers = {}\n- for path in remote_file_paths:\n- expected_headers.update(_make_headers(f\"{MODEL_URI}/{path}\"))\n- actual_headers = {}\n- for cred in creds:\n- actual_headers.update({header.name: header.value for header in cred.headers})\n- assert expected_headers == actual_headers\n-\n-\n-def test_download_from_cloud():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._get_download_presigned_url_and_headers\",\n- return_value=CreateDownloadUrlResponse(\n- url=_make_presigned_url(remote_file_path),\n- headers=[\n- HttpHeader(name=k, value=v) for k, v in _make_headers(remote_file_path).items()\n- ],\n- ),\n- ) as mock_request,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.download_file_using_http_uri\"\n- ) as mock_download,\n- ):\n- local_file = \"local_file\"\n- artifact_repo._download_from_cloud(remote_file_path, local_file)\n-\n- mock_request.assert_called_once_with(remote_file_path)\n- mock_download.assert_called_once_with(\n- http_uri=_make_presigned_url(remote_file_path),\n- download_path=local_file,\n- chunk_size=MLFLOW_MULTIPART_DOWNLOAD_CHUNK_SIZE.get(),\n- headers=_make_headers(remote_file_path),\n- )\n-\n-\n-def test_download_from_cloud_fail():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- endpoint, _ = FILESYSTEM_METHOD_TO_INFO[CreateDownloadUrlRequest]\n- exc_code = \"ENDPOINT_NOT_FOUND\"\n- exc_message = f\"Endpoint not found for {endpoint.lstrip('api')}.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with (\n- mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc),\n- pytest.raises(RestException) as exc_info, # noqa: PT011\n- ):\n- artifact_repo._download_from_cloud(remote_file_path, \"local_file\")\n-\n- assert exc_info.value.error_code == exc_code\n- assert str(exc_info.value) == f\"{exc_code}: {exc_message}\"\n-\n-\n-def test_log_artifact():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- local_file = \"local_file\"\n- artifact_path = \"remote/file/location\"\n- total_remote_path = f\"{artifact_path}/{os.path.basename(local_file)}\"\n- creds = ArtifactCredentialInfo(\n- signed_uri=_make_presigned_url(total_remote_path),\n- headers=[\n- ArtifactCredentialInfo.HttpHeader(name=k, value=v)\n- for k, v in _make_headers(total_remote_path).items()\n- ],\n- )\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._get_write_credential_infos\",\n- return_value=[creds],\n- ) as mock_request,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._upload_to_cloud\",\n- return_value=None,\n- ) as mock_upload,\n- ):\n- artifact_repo.log_artifact(local_file, artifact_path)\n- mock_request.assert_called_once_with(remote_file_paths=[total_remote_path])\n- mock_upload.assert_called_once_with(\n- cloud_credential_info=creds,\n- src_file_path=local_file,\n- artifact_file_path=total_remote_path,\n- )\n-\n-\n-def test_upload_to_cloud(tmp_path):\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- local_file = os.path.join(tmp_path, \"file.txt\")\n- content = \"content\"\n- with open(local_file, \"w\") as f:\n- f.write(content)\n- remote_file_path = \"some/remote/file/path\"\n- resp = mock.create_autospec(requests.Response, return_value=None)\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.cloud_storage_http_request\", return_value=resp\n- ) as mock_cloud,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.augmented_raise_for_status\"\n- ) as mock_status,\n- ):\n- cred_info = ArtifactCredentialInfo(\n- signed_uri=_make_presigned_url(remote_file_path),\n- headers=[\n- ArtifactCredentialInfo.HttpHeader(name=k, value=v)\n- for k, v in _make_headers(remote_file_path).items()\n- ],\n- )\n- artifact_repo._upload_to_cloud(cred_info, local_file, \"some/irrelevant/path\")\n- mock_cloud.assert_called_once_with(\n- \"put\",\n- _make_presigned_url(remote_file_path),\n- data=bytearray(content, \"utf-8\"),\n- headers=_make_headers(remote_file_path),\n- )\n- mock_status.assert_called_once_with(resp.__enter__())\n-\n-\n-def test_upload_to_cloud_fail():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- endpoint, _ = FILESYSTEM_METHOD_TO_INFO[CreateUploadUrlRequest]\n- exc_code = \"ENDPOINT_NOT_FOUND\"\n- exc_message = f\"Endpoint not found for {endpoint.lstrip('api')}.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with (\n- mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc),\n- pytest.raises(RestException) as exc_info, # noqa: PT011\n- ):\n- artifact_repo._download_from_cloud(remote_file_path, \"local_file\")\n-\n- assert exc_info.value.error_code == exc_code\n- assert str(exc_info.value) == f\"{exc_code}: {exc_message}\"\n-\n-\n-@pytest.mark.parametrize(\"throw\", [True, False])\n-@pytest.mark.parametrize(\"use_og_creds\", [True, False])\n-@pytest.mark.parametrize(\"status_code\", [401, 403])\n-def test_retry_refresh_creds_no_creds(throw, use_og_creds, status_code):\n- credentials = \"og_creds\"\n- called = False\n-\n- def creds_func():\n- nonlocal credentials\n- credentials = \"\".join(random.choices(string.ascii_lowercase, k=10))\n- return credentials\n-\n- def try_func(creds):\n- nonlocal called, credentials\n- assert creds == credentials\n- resp = requests.Response()\n- resp.status_code = status_code\n- if throw and not called:\n- called = True\n- raise requests.HTTPError(response=resp)\n-\n- mock_creds = mock.Mock(side_effect=creds_func)\n- mock_func = mock.Mock(side_effect=try_func)\n- if use_og_creds:\n- _retry_with_new_creds(try_func=mock_func, creds_func=mock_creds, orig_creds=credentials)\n- else:\n- _retry_with_new_creds(try_func=mock_func, creds_func=mock_creds)\n-\n- if throw:\n- assert mock_func.call_count == 2\n- assert mock_creds.call_count == 1 if use_og_creds else 2\n- else:\n- assert mock_func.call_count == 1\n- assert mock_creds.call_count == 0 if use_og_creds else 1\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S16\",\n \"shared_functions\": 1,\n \"shared_lines\": 15\n },\n {\n \"spec_id\": \"S30\",\n \"shared_functions\": 1,\n \"shared_lines\": 15\n },\n {\n \"spec_id\": \"S18\",\n \"shared_functions\": 1,\n \"shared_lines\": 8\n },\n {\n \"spec_id\": \"S26\",\n \"shared_functions\": 1,\n \"shared_lines\": 8\n },\n {\n \"spec_id\": \"S20\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S17\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n }\n ]\n}"}}},{"rowIdx":95,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_presigned_url_artifact_repo.8d57288e.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S21\",\n \"text\": \"Attempt to execute a function with credential retry logic for handling expired credentials.\\n\\nThis function implements a retry mechanism for operations that may fail due to credential expiration. It first attempts to execute the provided function with either the original credentials (if provided) or freshly generated credentials. If the operation fails with any exception, it assumes the credentials may have expired, generates new credentials, and retries the operation once.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/store/artifact/artifact_repo.py\",\n \"lines\": \"54-68\",\n \"function\": \"_retry_with_new_creds\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..e0f99f4b4 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..540e3ca5f 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,7 +157,7 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.proto_json_utils import parse_dict\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..4cd4ffc20 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,7 +39,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _UC_OSS_REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..53975a0af 100644\n--- a/mlflow/store/artifact/artifact_repo.py\n+++ b/mlflow/store/artifact/artifact_repo.py\n@@ -51,22 +51,22 @@ def _truncate_error(err: str, max_length: int = 10_000) -> str:\n return err[:half] + \"\\n\\n*** Error message is too long, truncated ***\\n\\n\" + err[-half:]\n \n \n-def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n- \"\"\"\n- Attempt the try_func with the original credentials (og_creds) if provided, or by generating the\n- credentials using creds_func. If the try_func throws, then try again with new credentials\n- provided by creds_func.\n- \"\"\"\n- try:\n- first_creds = creds_func() if orig_creds is None else orig_creds\n- return try_func(first_creds)\n- except Exception as e:\n- _logger.info(\n- f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n- \" Refreshing credentials and trying again...\"\n- )\n- new_creds = creds_func()\n- return try_func(new_creds)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -78,29 +78,29 @@ class ArtifactRepository:\n \n __metaclass__ = ABCMeta\n \n- def __init__(\n- self, artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None\n- ) -> None:\n- self.artifact_uri = artifact_uri\n- self.tracking_uri = tracking_uri\n- self.registry_uri = registry_uri\n- # Limit the number of threads used for artifact uploads/downloads. Use at most\n- # constants._NUM_MAX_THREADS threads or 2 * the number of CPU cores available on the\n- # system (whichever is smaller)\n- self.thread_pool = self._create_thread_pool()\n-\n- def log_artifact_handler(filename, artifact_path=None, artifact=None):\n- with tempfile.TemporaryDirectory() as tmp_dir:\n- tmp_path = os.path.join(tmp_dir, filename)\n- if artifact is not None:\n- # User should already have installed PIL to log a PIL image\n- from PIL import Image\n-\n- if isinstance(artifact, Image.Image):\n- artifact.save(tmp_path)\n- self.log_artifact(tmp_path, artifact_path)\n-\n- self._async_logging_queue = AsyncArtifactsLoggingQueue(log_artifact_handler)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __repr__(self) -> str:\n return (\n@@ -111,10 +111,10 @@ class ArtifactRepository:\n f\")\"\n )\n \n- def _create_thread_pool(self):\n- return ThreadPoolExecutor(\n- max_workers=self.max_workers, thread_name_prefix=f\"Mlflow{self.__class__.__name__}\"\n- )\n+\n+\n+\n+\n \n def flush_async_logging(self):\n \"\"\"\n@@ -355,11 +355,11 @@ class ArtifactRepository:\n artifact_path: Path of the artifact to delete.\n \"\"\"\n \n- @property\n- def max_workers(self) -> int:\n- \"\"\"Compute the number of workers to use for multi-threading.\"\"\"\n- num_cpus = os.cpu_count() or _NUM_DEFAULT_CPUS\n- return min(num_cpus * _NUM_MAX_THREADS_PER_CPU, _NUM_MAX_THREADS)\n+\n+\n+\n+\n+\n \n def download_trace_data(self) -> dict[str, Any]:\n \"\"\"\n@@ -473,4 +473,4 @@ def verify_artifact_path(artifact_path):\n if artifact_path and path_not_unique(artifact_path):\n raise MlflowException(\n f\"Invalid artifact path: '{artifact_path}'. {bad_path_message(artifact_path)}\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/store/artifact/azure_data_lake_artifact_repo.py b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\nindex 3d34f5971..b4d1dde04 100644\n--- a/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..2b13f2e8f 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,7 +62,7 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/gcs_artifact_repo.py b/mlflow/store/artifact/gcs_artifact_repo.py\nindex c68e7ac82..17febe6bb 100644\n--- a/mlflow/store/artifact/gcs_artifact_repo.py\n+++ b/mlflow/store/artifact/gcs_artifact_repo.py\n@@ -18,11 +18,11 @@ from mlflow.environment_variables import (\n MLFLOW_GCS_UPLOAD_CHUNK_SIZE,\n )\n from mlflow.exceptions import _UnsupportedMultipartUploadException\n-from mlflow.store.artifact.artifact_repo import (\n- ArtifactRepository,\n- MultipartUploadMixin,\n- _retry_with_new_creds,\n-)\n+from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+\n+\n+\n+\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n \n \ndiff --git a/mlflow/store/artifact/optimized_s3_artifact_repo.py b/mlflow/store/artifact/optimized_s3_artifact_repo.py\nindex f40f27abb..76b56ec82 100644\n--- a/mlflow/store/artifact/optimized_s3_artifact_repo.py\n+++ b/mlflow/store/artifact/optimized_s3_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..53db7931d 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -15,10 +15,10 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n ListDirectoryResponse,\n )\n from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..989d8bc69 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,7 +27,7 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..5e871abf6 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,7 +24,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _UC_OSS_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..2a76bb66f 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,7 +37,7 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..da41b2531 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,7 +52,7 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V4_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..4b3652f73 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,7 +120,7 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V3_TRACE_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..f37d92aa6 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,7 +43,7 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/utils/async_logging/async_artifacts_logging_queue.py b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\nindex 08c89366a..87a54662a 100644\n--- a/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n+++ b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n@@ -31,15 +31,15 @@ class AsyncArtifactsLoggingQueue:\n - artifact: The artifact to be logged.\n \"\"\"\n \n- def __init__(\n- self, artifact_logging_func: Callable[[str, str, Union[\"PIL.Image.Image\"]], None]\n- ) -> None:\n- self._queue: Queue[RunArtifact] = Queue()\n- self._lock = threading.RLock()\n- self._artifact_logging_func = artifact_logging_func\n \n- self._stop_data_logging_thread_event = threading.Event()\n- self._is_activated = False\n+\n+\n+\n+\n+\n+\n+\n+\n \n def _at_exit_callback(self) -> None:\n \"\"\"Callback function to be executed when the program is exiting.\n@@ -255,4 +255,4 @@ class AsyncArtifactsLoggingQueue:\n self._set_up_logging_thread()\n atexit.register(self._at_exit_callback)\n \n- self._is_activated = True\n+ self._is_activated = True\n\\ No newline at end of file\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..61480cb97 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -44,89 +44,89 @@ def _mark_int64_fields_for_proto_maps(proto_map, value_field_type):\n return json_dict\n \n \n-def _mark_int64_fields(proto_message):\n- \"\"\"Converts a proto message to JSON, preserving only int64-related fields.\"\"\"\n- json_dict = {}\n- for field, value in proto_message.ListFields():\n- if (\n- # These three conditions check if this field is a protobuf map.\n- # See the official implementation: https://bit.ly/3EMx1rl\n- field.type == FieldDescriptor.TYPE_MESSAGE\n- and field.message_type.has_options\n- and field.message_type.GetOptions().map_entry\n- ):\n- # Deal with proto map fields separately in another function.\n- json_dict[field.name] = _mark_int64_fields_for_proto_maps(\n- value, field.message_type.fields_by_name[\"value\"].type\n- )\n- continue\n \n- if field.type == FieldDescriptor.TYPE_MESSAGE:\n- ftype = partial(_mark_int64_fields)\n- elif field.type in _PROTOBUF_INT64_FIELDS:\n- ftype = int\n- else:\n- # Skip all non-int64 fields.\n- continue\n \n- # Use is_repeated property (preferred) with fallback to deprecated label\n- try:\n- is_repeated = field.is_repeated\n- except AttributeError:\n- is_repeated = field.label == FieldDescriptor.LABEL_REPEATED\n \n- json_dict[field.name] = [ftype(v) for v in value] if is_repeated else ftype(value)\n- return json_dict\n \n \n-def _merge_json_dicts(from_dict, to_dict):\n- \"\"\"Merges the json elements of from_dict into to_dict. Only works for json dicts\n- converted from proto messages\n- \"\"\"\n- for key, value in from_dict.items():\n- if isinstance(key, int) and str(key) in to_dict:\n- # When the key (i.e. the proto field name) is an integer, it must be a proto map field\n- # with integer as the key. For example:\n- # from_dict is {'field_map': {1: '2', 3: '4'}}\n- # to_dict is {'field_map': {'1': '2', '3': '4'}}\n- # So we need to replace the str keys with int keys in to_dict.\n- to_dict[key] = to_dict[str(key)]\n- del to_dict[str(key)]\n-\n- if key not in to_dict:\n- continue\n-\n- if isinstance(value, dict):\n- _merge_json_dicts(from_dict[key], to_dict[key])\n- elif isinstance(value, list):\n- for i, v in enumerate(value):\n- if isinstance(v, dict):\n- _merge_json_dicts(v, to_dict[key][i])\n- else:\n- to_dict[key][i] = v\n- else:\n- to_dict[key] = from_dict[key]\n- return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py\nindex 5d5de9527..aa5b3c40a 100644\n--- a/tests/store/artifact/test_presigned_url_artifact_repo.py\n+++ b/tests/store/artifact/test_presigned_url_artifact_repo.py\n@@ -20,13 +20,13 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n HttpHeader,\n ListDirectoryResponse,\n )\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.presigned_url_artifact_repo import (\n DIRECTORIES_ENDPOINT,\n FILESYSTEM_METHOD_TO_INFO,\n PresignedUrlArtifactRepository,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n \n MODEL_NAME = \"catalog.schema.model\"\n MODEL_VERSION = 1\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..e0f99f4b4 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..540e3ca5f 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,7 +157,7 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.proto_json_utils import parse_dict\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..4cd4ffc20 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,7 +39,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _UC_OSS_REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..53975a0af 100644\n+++ a/mlflow/store/artifact/artifact_repo.py\n--- b/mlflow/store/artifact/artifact_repo.py\n@@ -51,22 +51,22 @@ def _truncate_error(err: str, max_length: int = 10_000) -> str:\n return err[:half] + \"\\n\\n*** Error message is too long, truncated ***\\n\\n\" + err[-half:]\n \n \n+def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n+ \"\"\"\n+ Attempt the try_func with the original credentials (og_creds) if provided, or by generating the\n+ credentials using creds_func. If the try_func throws, then try again with new credentials\n+ provided by creds_func.\n+ \"\"\"\n+ try:\n+ first_creds = creds_func() if orig_creds is None else orig_creds\n+ return try_func(first_creds)\n+ except Exception as e:\n+ _logger.info(\n+ f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n+ \" Refreshing credentials and trying again...\"\n+ )\n+ new_creds = creds_func()\n+ return try_func(new_creds)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -78,29 +78,29 @@ class ArtifactRepository:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(\n+ self, artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None\n+ ) -> None:\n+ self.artifact_uri = artifact_uri\n+ self.tracking_uri = tracking_uri\n+ self.registry_uri = registry_uri\n+ # Limit the number of threads used for artifact uploads/downloads. Use at most\n+ # constants._NUM_MAX_THREADS threads or 2 * the number of CPU cores available on the\n+ # system (whichever is smaller)\n+ self.thread_pool = self._create_thread_pool()\n+\n+ def log_artifact_handler(filename, artifact_path=None, artifact=None):\n+ with tempfile.TemporaryDirectory() as tmp_dir:\n+ tmp_path = os.path.join(tmp_dir, filename)\n+ if artifact is not None:\n+ # User should already have installed PIL to log a PIL image\n+ from PIL import Image\n+\n+ if isinstance(artifact, Image.Image):\n+ artifact.save(tmp_path)\n+ self.log_artifact(tmp_path, artifact_path)\n+\n+ self._async_logging_queue = AsyncArtifactsLoggingQueue(log_artifact_handler)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __repr__(self) -> str:\n return (\n@@ -111,10 +111,10 @@ class ArtifactRepository:\n f\")\"\n )\n \n+ def _create_thread_pool(self):\n+ return ThreadPoolExecutor(\n+ max_workers=self.max_workers, thread_name_prefix=f\"Mlflow{self.__class__.__name__}\"\n+ )\n-\n-\n-\n-\n \n def flush_async_logging(self):\n \"\"\"\n@@ -355,11 +355,11 @@ class ArtifactRepository:\n artifact_path: Path of the artifact to delete.\n \"\"\"\n \n+ @property\n+ def max_workers(self) -> int:\n+ \"\"\"Compute the number of workers to use for multi-threading.\"\"\"\n+ num_cpus = os.cpu_count() or _NUM_DEFAULT_CPUS\n+ return min(num_cpus * _NUM_MAX_THREADS_PER_CPU, _NUM_MAX_THREADS)\n-\n-\n-\n-\n-\n \n def download_trace_data(self) -> dict[str, Any]:\n \"\"\"\n@@ -473,4 +473,4 @@ def verify_artifact_path(artifact_path):\n if artifact_path and path_not_unique(artifact_path):\n raise MlflowException(\n f\"Invalid artifact path: '{artifact_path}'. {bad_path_message(artifact_path)}\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/store/artifact/azure_data_lake_artifact_repo.py b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\nindex 3d34f5971..b4d1dde04 100644\n+++ a/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n--- b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..2b13f2e8f 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,7 +62,7 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/gcs_artifact_repo.py b/mlflow/store/artifact/gcs_artifact_repo.py\nindex c68e7ac82..17febe6bb 100644\n+++ a/mlflow/store/artifact/gcs_artifact_repo.py\n--- b/mlflow/store/artifact/gcs_artifact_repo.py\n@@ -18,11 +18,11 @@ from mlflow.environment_variables import (\n MLFLOW_GCS_UPLOAD_CHUNK_SIZE,\n )\n from mlflow.exceptions import _UnsupportedMultipartUploadException\n+from mlflow.store.artifact.artifact_repo import (\n+ ArtifactRepository,\n+ MultipartUploadMixin,\n+ _retry_with_new_creds,\n+)\n-from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-\n-\n-\n-\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n \n \ndiff --git a/mlflow/store/artifact/optimized_s3_artifact_repo.py b/mlflow/store/artifact/optimized_s3_artifact_repo.py\nindex f40f27abb..76b56ec82 100644\n+++ a/mlflow/store/artifact/optimized_s3_artifact_repo.py\n--- b/mlflow/store/artifact/optimized_s3_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..53db7931d 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -15,10 +15,10 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n ListDirectoryResponse,\n )\n from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..989d8bc69 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,7 +27,7 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..5e871abf6 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,7 +24,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _UC_OSS_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..2a76bb66f 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,7 +37,7 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..da41b2531 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,7 +52,7 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V4_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..4b3652f73 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,7 +120,7 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V3_TRACE_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..f37d92aa6 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,7 +43,7 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/utils/async_logging/async_artifacts_logging_queue.py b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\nindex 08c89366a..87a54662a 100644\n+++ a/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n--- b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n@@ -31,15 +31,15 @@ class AsyncArtifactsLoggingQueue:\n - artifact: The artifact to be logged.\n \"\"\"\n \n+ def __init__(\n+ self, artifact_logging_func: Callable[[str, str, Union[\"PIL.Image.Image\"]], None]\n+ ) -> None:\n+ self._queue: Queue[RunArtifact] = Queue()\n+ self._lock = threading.RLock()\n+ self._artifact_logging_func = artifact_logging_func\n \n+ self._stop_data_logging_thread_event = threading.Event()\n+ self._is_activated = False\n-\n-\n-\n-\n-\n-\n-\n-\n \n def _at_exit_callback(self) -> None:\n \"\"\"Callback function to be executed when the program is exiting.\n@@ -255,4 +255,4 @@ class AsyncArtifactsLoggingQueue:\n self._set_up_logging_thread()\n atexit.register(self._at_exit_callback)\n \n+ self._is_activated = True\n- self._is_activated = True\n\\ No newline at end of file\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..61480cb97 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -44,89 +44,89 @@ def _mark_int64_fields_for_proto_maps(proto_map, value_field_type):\n return json_dict\n \n \n+def _mark_int64_fields(proto_message):\n+ \"\"\"Converts a proto message to JSON, preserving only int64-related fields.\"\"\"\n+ json_dict = {}\n+ for field, value in proto_message.ListFields():\n+ if (\n+ # These three conditions check if this field is a protobuf map.\n+ # See the official implementation: https://bit.ly/3EMx1rl\n+ field.type == FieldDescriptor.TYPE_MESSAGE\n+ and field.message_type.has_options\n+ and field.message_type.GetOptions().map_entry\n+ ):\n+ # Deal with proto map fields separately in another function.\n+ json_dict[field.name] = _mark_int64_fields_for_proto_maps(\n+ value, field.message_type.fields_by_name[\"value\"].type\n+ )\n+ continue\n \n+ if field.type == FieldDescriptor.TYPE_MESSAGE:\n+ ftype = partial(_mark_int64_fields)\n+ elif field.type in _PROTOBUF_INT64_FIELDS:\n+ ftype = int\n+ else:\n+ # Skip all non-int64 fields.\n+ continue\n \n+ # Use is_repeated property (preferred) with fallback to deprecated label\n+ try:\n+ is_repeated = field.is_repeated\n+ except AttributeError:\n+ is_repeated = field.label == FieldDescriptor.LABEL_REPEATED\n \n+ json_dict[field.name] = [ftype(v) for v in value] if is_repeated else ftype(value)\n+ return json_dict\n \n \n+def _merge_json_dicts(from_dict, to_dict):\n+ \"\"\"Merges the json elements of from_dict into to_dict. Only works for json dicts\n+ converted from proto messages\n+ \"\"\"\n+ for key, value in from_dict.items():\n+ if isinstance(key, int) and str(key) in to_dict:\n+ # When the key (i.e. the proto field name) is an integer, it must be a proto map field\n+ # with integer as the key. For example:\n+ # from_dict is {'field_map': {1: '2', 3: '4'}}\n+ # to_dict is {'field_map': {'1': '2', '3': '4'}}\n+ # So we need to replace the str keys with int keys in to_dict.\n+ to_dict[key] = to_dict[str(key)]\n+ del to_dict[str(key)]\n+\n+ if key not in to_dict:\n+ continue\n+\n+ if isinstance(value, dict):\n+ _merge_json_dicts(from_dict[key], to_dict[key])\n+ elif isinstance(value, list):\n+ for i, v in enumerate(value):\n+ if isinstance(v, dict):\n+ _merge_json_dicts(v, to_dict[key][i])\n+ else:\n+ to_dict[key][i] = v\n+ else:\n+ to_dict[key] = from_dict[key]\n+ return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..dd8f4e6a6 100644\n--- a/mlflow/store/artifact/artifact_repo.py\n+++ b/mlflow/store/artifact/artifact_repo.py\n@@ -58,7 +58,7 @@ def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n provided by creds_func.\n \"\"\"\n try:\n- first_creds = creds_func() if orig_creds is None else orig_creds\n+ first_creds = creds_func() if not orig_creds else orig_creds\n return try_func(first_creds)\n except Exception as e:\n _logger.info(\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py\ndeleted file mode 100644\nindex 5d5de9527..000000000\n--- a/tests/store/artifact/test_presigned_url_artifact_repo.py\n+++ /dev/null\n@@ -1,363 +0,0 @@\n-import json\n-import os\n-import random\n-import string\n-from unittest import mock\n-from unittest.mock import ANY\n-\n-import pytest\n-import requests\n-\n-from mlflow.environment_variables import MLFLOW_MULTIPART_DOWNLOAD_CHUNK_SIZE\n-from mlflow.exceptions import RestException\n-from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.protos.databricks_filesystem_service_pb2 import (\n- CreateDownloadUrlRequest,\n- CreateDownloadUrlResponse,\n- CreateUploadUrlRequest,\n- CreateUploadUrlResponse,\n- DirectoryEntry,\n- HttpHeader,\n- ListDirectoryResponse,\n-)\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-from mlflow.store.artifact.presigned_url_artifact_repo import (\n- DIRECTORIES_ENDPOINT,\n- FILESYSTEM_METHOD_TO_INFO,\n- PresignedUrlArtifactRepository,\n-)\n-from mlflow.utils.proto_json_utils import message_to_json\n-\n-MODEL_NAME = \"catalog.schema.model\"\n-MODEL_VERSION = 1\n-MODEL_URI = \"/Models/catalog/schema/model/1\"\n-PRESIGNED_URL_ARTIFACT_REPOSITORY = \"mlflow.store.artifact.presigned_url_artifact_repo\"\n-_DATABRICKS_UC_SCHEME = \"databricks-uc\"\n-\n-\n-@pytest.fixture(autouse=True)\n-def run_around_tests():\n- # mock this call to credentials for all tests in suite\n- with mock.patch(\"mlflow.utils.databricks_utils.get_databricks_host_creds\"):\n- yield\n-\n-\n-def test_artifact_uri():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- assert MODEL_URI == artifact_repo.artifact_uri\n-\n-\n-def mock_list_directory(*args, **kwargs):\n- endpoint = kwargs[\"endpoint\"]\n- json_body = kwargs[\"json_body\"]\n-\n- if endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/dir\" and json_body == json.dumps(\n- {\"page_token\": \"some_token\"}\n- ):\n- return ListDirectoryResponse(\n- contents=[\n- DirectoryEntry(is_directory=False, path=f\"{MODEL_URI}/dir/file2\", file_size=2)\n- ]\n- )\n- elif endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/dir\":\n- return ListDirectoryResponse(\n- contents=[\n- DirectoryEntry(is_directory=False, path=f\"{MODEL_URI}/dir/file1\", file_size=1)\n- ],\n- next_page_token=\"some_token\",\n- )\n- elif endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/\":\n- return ListDirectoryResponse(\n- contents=[DirectoryEntry(is_directory=True, path=f\"{MODEL_URI}/dir\")],\n- )\n- else:\n- raise ValueError(f\"Unexpected endpoint: {endpoint}\")\n-\n-\n-def test_list_artifact_pagination():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_list_directory\n- ) as mock_list:\n- resp = artifact_repo.list_artifacts()\n- assert len(resp) == 1\n- assert resp[0].path == \"dir\"\n- assert resp[0].is_dir is True\n- assert resp[0].file_size is None\n-\n- resp = artifact_repo.list_artifacts(\"dir\")\n- assert len(resp) == 2\n- assert {r.path for r in resp} == {\"dir/file1\", \"dir/file2\"}\n- assert {r.is_dir for r in resp} == {False}\n- assert {r.file_size for r in resp} == {1, 2}\n-\n- assert mock_list.call_count == 3\n-\n-\n-def test_list_artifacts_failure():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- exc_code = \"NOT_FOUND\"\n- exc_message = \"The directory being accessed is not found.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc):\n- empty_infos = artifact_repo.list_artifacts(remote_file_path)\n- assert len(empty_infos) == 0\n-\n-\n-def _make_presigned_url(remote_path):\n- return f\"presigned_url/{remote_path}\"\n-\n-\n-def _make_headers(remote_path):\n- return {f\"path_header-{remote_path}\": f\"remote-path={remote_path}\"}\n-\n-\n-def mock_create_download_url(*args, **kwargs):\n- remote_path = json.loads(kwargs[\"json_body\"])[\"path\"]\n- return CreateDownloadUrlResponse(\n- url=_make_presigned_url(remote_path),\n- headers=[\n- HttpHeader(name=header, value=val) for header, val in _make_headers(remote_path).items()\n- ],\n- )\n-\n-\n-def test_get_read_credentials():\n- remote_file_paths = [\"file\", \"dir/file1\", \"dir/file2\"]\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- download_url_endpoint = \"/api/2.0/fs/create-download-url\"\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_create_download_url\n- ) as mock_call_endpoint:\n- creds = artifact_repo._get_read_credential_infos(remote_file_paths)\n- assert mock_call_endpoint.call_count == 3\n-\n- for remote_file_path in remote_file_paths:\n- mock_call_endpoint.assert_any_call(\n- host_creds=ANY,\n- endpoint=f\"{download_url_endpoint}\",\n- method=\"POST\",\n- json_body=message_to_json(\n- CreateDownloadUrlRequest(path=f\"{MODEL_URI}/{remote_file_path}\")\n- ),\n- response_proto=ANY,\n- )\n-\n- assert {_make_presigned_url(f\"{MODEL_URI}/{path}\") for path in remote_file_paths} == {\n- cred.signed_uri for cred in creds\n- }\n- expected_headers = {}\n- for path in remote_file_paths:\n- expected_headers.update(_make_headers(f\"{MODEL_URI}/{path}\"))\n- actual_headers = {}\n- for cred in creds:\n- actual_headers.update({header.name: header.value for header in cred.headers})\n- assert expected_headers == actual_headers\n-\n-\n-def mock_create_upload_url(*args, **kwargs):\n- remote_path = json.loads(kwargs[\"json_body\"])[\"path\"]\n- return CreateUploadUrlResponse(\n- url=_make_presigned_url(remote_path),\n- headers=[\n- HttpHeader(name=header, value=val) for header, val in _make_headers(remote_path).items()\n- ],\n- )\n-\n-\n-def test_get_write_credentials():\n- remote_file_paths = [\"file\", \"dir/file1\", \"dir/file2\"]\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- upload_url_endpoint = \"/api/2.0/fs/create-upload-url\"\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_create_upload_url\n- ) as mock_call_endpoint:\n- creds = artifact_repo._get_write_credential_infos(remote_file_paths)\n- assert mock_call_endpoint.call_count == 3\n- for remote_file_path in remote_file_paths:\n- mock_call_endpoint.assert_any_call(\n- host_creds=ANY,\n- endpoint=f\"{upload_url_endpoint}\",\n- method=\"POST\",\n- json_body=message_to_json(\n- CreateUploadUrlRequest(path=f\"{MODEL_URI}/{remote_file_path}\")\n- ),\n- response_proto=ANY,\n- )\n-\n- expected_headers = {}\n- for path in remote_file_paths:\n- expected_headers.update(_make_headers(f\"{MODEL_URI}/{path}\"))\n- actual_headers = {}\n- for cred in creds:\n- actual_headers.update({header.name: header.value for header in cred.headers})\n- assert expected_headers == actual_headers\n-\n-\n-def test_download_from_cloud():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._get_download_presigned_url_and_headers\",\n- return_value=CreateDownloadUrlResponse(\n- url=_make_presigned_url(remote_file_path),\n- headers=[\n- HttpHeader(name=k, value=v) for k, v in _make_headers(remote_file_path).items()\n- ],\n- ),\n- ) as mock_request,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.download_file_using_http_uri\"\n- ) as mock_download,\n- ):\n- local_file = \"local_file\"\n- artifact_repo._download_from_cloud(remote_file_path, local_file)\n-\n- mock_request.assert_called_once_with(remote_file_path)\n- mock_download.assert_called_once_with(\n- http_uri=_make_presigned_url(remote_file_path),\n- download_path=local_file,\n- chunk_size=MLFLOW_MULTIPART_DOWNLOAD_CHUNK_SIZE.get(),\n- headers=_make_headers(remote_file_path),\n- )\n-\n-\n-def test_download_from_cloud_fail():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- endpoint, _ = FILESYSTEM_METHOD_TO_INFO[CreateDownloadUrlRequest]\n- exc_code = \"ENDPOINT_NOT_FOUND\"\n- exc_message = f\"Endpoint not found for {endpoint.lstrip('api')}.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with (\n- mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc),\n- pytest.raises(RestException) as exc_info, # noqa: PT011\n- ):\n- artifact_repo._download_from_cloud(remote_file_path, \"local_file\")\n-\n- assert exc_info.value.error_code == exc_code\n- assert str(exc_info.value) == f\"{exc_code}: {exc_message}\"\n-\n-\n-def test_log_artifact():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- local_file = \"local_file\"\n- artifact_path = \"remote/file/location\"\n- total_remote_path = f\"{artifact_path}/{os.path.basename(local_file)}\"\n- creds = ArtifactCredentialInfo(\n- signed_uri=_make_presigned_url(total_remote_path),\n- headers=[\n- ArtifactCredentialInfo.HttpHeader(name=k, value=v)\n- for k, v in _make_headers(total_remote_path).items()\n- ],\n- )\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._get_write_credential_infos\",\n- return_value=[creds],\n- ) as mock_request,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._upload_to_cloud\",\n- return_value=None,\n- ) as mock_upload,\n- ):\n- artifact_repo.log_artifact(local_file, artifact_path)\n- mock_request.assert_called_once_with(remote_file_paths=[total_remote_path])\n- mock_upload.assert_called_once_with(\n- cloud_credential_info=creds,\n- src_file_path=local_file,\n- artifact_file_path=total_remote_path,\n- )\n-\n-\n-def test_upload_to_cloud(tmp_path):\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- local_file = os.path.join(tmp_path, \"file.txt\")\n- content = \"content\"\n- with open(local_file, \"w\") as f:\n- f.write(content)\n- remote_file_path = \"some/remote/file/path\"\n- resp = mock.create_autospec(requests.Response, return_value=None)\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.cloud_storage_http_request\", return_value=resp\n- ) as mock_cloud,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.augmented_raise_for_status\"\n- ) as mock_status,\n- ):\n- cred_info = ArtifactCredentialInfo(\n- signed_uri=_make_presigned_url(remote_file_path),\n- headers=[\n- ArtifactCredentialInfo.HttpHeader(name=k, value=v)\n- for k, v in _make_headers(remote_file_path).items()\n- ],\n- )\n- artifact_repo._upload_to_cloud(cred_info, local_file, \"some/irrelevant/path\")\n- mock_cloud.assert_called_once_with(\n- \"put\",\n- _make_presigned_url(remote_file_path),\n- data=bytearray(content, \"utf-8\"),\n- headers=_make_headers(remote_file_path),\n- )\n- mock_status.assert_called_once_with(resp.__enter__())\n-\n-\n-def test_upload_to_cloud_fail():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- endpoint, _ = FILESYSTEM_METHOD_TO_INFO[CreateUploadUrlRequest]\n- exc_code = \"ENDPOINT_NOT_FOUND\"\n- exc_message = f\"Endpoint not found for {endpoint.lstrip('api')}.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with (\n- mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc),\n- pytest.raises(RestException) as exc_info, # noqa: PT011\n- ):\n- artifact_repo._download_from_cloud(remote_file_path, \"local_file\")\n-\n- assert exc_info.value.error_code == exc_code\n- assert str(exc_info.value) == f\"{exc_code}: {exc_message}\"\n-\n-\n-@pytest.mark.parametrize(\"throw\", [True, False])\n-@pytest.mark.parametrize(\"use_og_creds\", [True, False])\n-@pytest.mark.parametrize(\"status_code\", [401, 403])\n-def test_retry_refresh_creds_no_creds(throw, use_og_creds, status_code):\n- credentials = \"og_creds\"\n- called = False\n-\n- def creds_func():\n- nonlocal credentials\n- credentials = \"\".join(random.choices(string.ascii_lowercase, k=10))\n- return credentials\n-\n- def try_func(creds):\n- nonlocal called, credentials\n- assert creds == credentials\n- resp = requests.Response()\n- resp.status_code = status_code\n- if throw and not called:\n- called = True\n- raise requests.HTTPError(response=resp)\n-\n- mock_creds = mock.Mock(side_effect=creds_func)\n- mock_func = mock.Mock(side_effect=try_func)\n- if use_og_creds:\n- _retry_with_new_creds(try_func=mock_func, creds_func=mock_creds, orig_creds=credentials)\n- else:\n- _retry_with_new_creds(try_func=mock_func, creds_func=mock_creds)\n-\n- if throw:\n- assert mock_func.call_count == 2\n- assert mock_creds.call_count == 1 if use_og_creds else 2\n- else:\n- assert mock_func.call_count == 1\n- assert mock_creds.call_count == 0 if use_og_creds else 1\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S23\",\n \"shared_functions\": 1,\n \"shared_lines\": 6\n },\n {\n \"spec_id\": \"S24\",\n \"shared_functions\": 1,\n \"shared_lines\": 6\n },\n {\n \"spec_id\": \"S22\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S25\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n }\n ]\n}"}}},{"rowIdx":96,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_presigned_url_artifact_repo.8d57288e.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S22\",\n \"text\": \"Returns:\\n The return value of try_func when it executes successfully.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/store/artifact/artifact_repo.py\",\n \"lines\": \"62\",\n \"function\": \"_retry_with_new_creds\"\n },\n {\n \"file\": \"mlflow/store/artifact/artifact_repo.py\",\n \"lines\": \"68\",\n \"function\": \"_retry_with_new_creds\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..e0f99f4b4 100644\n--- a/mlflow/data/delta_dataset_source.py\n+++ b/mlflow/data/delta_dataset_source.py\n@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..540e3ca5f 100644\n--- a/mlflow/store/_unity_catalog/registry/rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,7 +157,7 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n-from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n+from mlflow.utils.proto_json_utils import parse_dict\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..4cd4ffc20 100644\n--- a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n+++ b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,7 +39,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _UC_OSS_REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..53975a0af 100644\n--- a/mlflow/store/artifact/artifact_repo.py\n+++ b/mlflow/store/artifact/artifact_repo.py\n@@ -51,22 +51,22 @@ def _truncate_error(err: str, max_length: int = 10_000) -> str:\n return err[:half] + \"\\n\\n*** Error message is too long, truncated ***\\n\\n\" + err[-half:]\n \n \n-def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n- \"\"\"\n- Attempt the try_func with the original credentials (og_creds) if provided, or by generating the\n- credentials using creds_func. If the try_func throws, then try again with new credentials\n- provided by creds_func.\n- \"\"\"\n- try:\n- first_creds = creds_func() if orig_creds is None else orig_creds\n- return try_func(first_creds)\n- except Exception as e:\n- _logger.info(\n- f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n- \" Refreshing credentials and trying again...\"\n- )\n- new_creds = creds_func()\n- return try_func(new_creds)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @developer_stable\n@@ -78,29 +78,29 @@ class ArtifactRepository:\n \n __metaclass__ = ABCMeta\n \n- def __init__(\n- self, artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None\n- ) -> None:\n- self.artifact_uri = artifact_uri\n- self.tracking_uri = tracking_uri\n- self.registry_uri = registry_uri\n- # Limit the number of threads used for artifact uploads/downloads. Use at most\n- # constants._NUM_MAX_THREADS threads or 2 * the number of CPU cores available on the\n- # system (whichever is smaller)\n- self.thread_pool = self._create_thread_pool()\n-\n- def log_artifact_handler(filename, artifact_path=None, artifact=None):\n- with tempfile.TemporaryDirectory() as tmp_dir:\n- tmp_path = os.path.join(tmp_dir, filename)\n- if artifact is not None:\n- # User should already have installed PIL to log a PIL image\n- from PIL import Image\n-\n- if isinstance(artifact, Image.Image):\n- artifact.save(tmp_path)\n- self.log_artifact(tmp_path, artifact_path)\n-\n- self._async_logging_queue = AsyncArtifactsLoggingQueue(log_artifact_handler)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def __repr__(self) -> str:\n return (\n@@ -111,10 +111,10 @@ class ArtifactRepository:\n f\")\"\n )\n \n- def _create_thread_pool(self):\n- return ThreadPoolExecutor(\n- max_workers=self.max_workers, thread_name_prefix=f\"Mlflow{self.__class__.__name__}\"\n- )\n+\n+\n+\n+\n \n def flush_async_logging(self):\n \"\"\"\n@@ -355,11 +355,11 @@ class ArtifactRepository:\n artifact_path: Path of the artifact to delete.\n \"\"\"\n \n- @property\n- def max_workers(self) -> int:\n- \"\"\"Compute the number of workers to use for multi-threading.\"\"\"\n- num_cpus = os.cpu_count() or _NUM_DEFAULT_CPUS\n- return min(num_cpus * _NUM_MAX_THREADS_PER_CPU, _NUM_MAX_THREADS)\n+\n+\n+\n+\n+\n \n def download_trace_data(self) -> dict[str, Any]:\n \"\"\"\n@@ -473,4 +473,4 @@ def verify_artifact_path(artifact_path):\n if artifact_path and path_not_unique(artifact_path):\n raise MlflowException(\n f\"Invalid artifact path: '{artifact_path}'. {bad_path_message(artifact_path)}\"\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/store/artifact/azure_data_lake_artifact_repo.py b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\nindex 3d34f5971..b4d1dde04 100644\n--- a/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n+++ b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..2b13f2e8f 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,7 +62,7 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n--- a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n+++ b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/gcs_artifact_repo.py b/mlflow/store/artifact/gcs_artifact_repo.py\nindex c68e7ac82..17febe6bb 100644\n--- a/mlflow/store/artifact/gcs_artifact_repo.py\n+++ b/mlflow/store/artifact/gcs_artifact_repo.py\n@@ -18,11 +18,11 @@ from mlflow.environment_variables import (\n MLFLOW_GCS_UPLOAD_CHUNK_SIZE,\n )\n from mlflow.exceptions import _UnsupportedMultipartUploadException\n-from mlflow.store.artifact.artifact_repo import (\n- ArtifactRepository,\n- MultipartUploadMixin,\n- _retry_with_new_creds,\n-)\n+from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n+\n+\n+\n+\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n \n \ndiff --git a/mlflow/store/artifact/optimized_s3_artifact_repo.py b/mlflow/store/artifact/optimized_s3_artifact_repo.py\nindex f40f27abb..76b56ec82 100644\n--- a/mlflow/store/artifact/optimized_s3_artifact_repo.py\n+++ b/mlflow/store/artifact/optimized_s3_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..53db7931d 100644\n--- a/mlflow/store/artifact/presigned_url_artifact_repo.py\n+++ b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -15,10 +15,10 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n ListDirectoryResponse,\n )\n from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..989d8bc69 100644\n--- a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,7 +27,7 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..5e871abf6 100644\n--- a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n+++ b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,7 +24,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _UC_OSS_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..2a76bb66f 100644\n--- a/mlflow/store/model_registry/rest_store.py\n+++ b/mlflow/store/model_registry/rest_store.py\n@@ -37,7 +37,7 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..da41b2531 100644\n--- a/mlflow/store/tracking/databricks_rest_store.py\n+++ b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,7 +52,7 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V4_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..4b3652f73 100644\n--- a/mlflow/store/tracking/rest_store.py\n+++ b/mlflow/store/tracking/rest_store.py\n@@ -120,7 +120,7 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V3_TRACE_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..f37d92aa6 100644\n--- a/mlflow/utils/_unity_catalog_utils.py\n+++ b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,7 +43,7 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/utils/async_logging/async_artifacts_logging_queue.py b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\nindex 08c89366a..87a54662a 100644\n--- a/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n+++ b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n@@ -31,15 +31,15 @@ class AsyncArtifactsLoggingQueue:\n - artifact: The artifact to be logged.\n \"\"\"\n \n- def __init__(\n- self, artifact_logging_func: Callable[[str, str, Union[\"PIL.Image.Image\"]], None]\n- ) -> None:\n- self._queue: Queue[RunArtifact] = Queue()\n- self._lock = threading.RLock()\n- self._artifact_logging_func = artifact_logging_func\n \n- self._stop_data_logging_thread_event = threading.Event()\n- self._is_activated = False\n+\n+\n+\n+\n+\n+\n+\n+\n \n def _at_exit_callback(self) -> None:\n \"\"\"Callback function to be executed when the program is exiting.\n@@ -255,4 +255,4 @@ class AsyncArtifactsLoggingQueue:\n self._set_up_logging_thread()\n atexit.register(self._at_exit_callback)\n \n- self._is_activated = True\n+ self._is_activated = True\n\\ No newline at end of file\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..61480cb97 100644\n--- a/mlflow/utils/proto_json_utils.py\n+++ b/mlflow/utils/proto_json_utils.py\n@@ -44,89 +44,89 @@ def _mark_int64_fields_for_proto_maps(proto_map, value_field_type):\n return json_dict\n \n \n-def _mark_int64_fields(proto_message):\n- \"\"\"Converts a proto message to JSON, preserving only int64-related fields.\"\"\"\n- json_dict = {}\n- for field, value in proto_message.ListFields():\n- if (\n- # These three conditions check if this field is a protobuf map.\n- # See the official implementation: https://bit.ly/3EMx1rl\n- field.type == FieldDescriptor.TYPE_MESSAGE\n- and field.message_type.has_options\n- and field.message_type.GetOptions().map_entry\n- ):\n- # Deal with proto map fields separately in another function.\n- json_dict[field.name] = _mark_int64_fields_for_proto_maps(\n- value, field.message_type.fields_by_name[\"value\"].type\n- )\n- continue\n \n- if field.type == FieldDescriptor.TYPE_MESSAGE:\n- ftype = partial(_mark_int64_fields)\n- elif field.type in _PROTOBUF_INT64_FIELDS:\n- ftype = int\n- else:\n- # Skip all non-int64 fields.\n- continue\n \n- # Use is_repeated property (preferred) with fallback to deprecated label\n- try:\n- is_repeated = field.is_repeated\n- except AttributeError:\n- is_repeated = field.label == FieldDescriptor.LABEL_REPEATED\n \n- json_dict[field.name] = [ftype(v) for v in value] if is_repeated else ftype(value)\n- return json_dict\n \n \n-def _merge_json_dicts(from_dict, to_dict):\n- \"\"\"Merges the json elements of from_dict into to_dict. Only works for json dicts\n- converted from proto messages\n- \"\"\"\n- for key, value in from_dict.items():\n- if isinstance(key, int) and str(key) in to_dict:\n- # When the key (i.e. the proto field name) is an integer, it must be a proto map field\n- # with integer as the key. For example:\n- # from_dict is {'field_map': {1: '2', 3: '4'}}\n- # to_dict is {'field_map': {'1': '2', '3': '4'}}\n- # So we need to replace the str keys with int keys in to_dict.\n- to_dict[key] = to_dict[str(key)]\n- del to_dict[str(key)]\n-\n- if key not in to_dict:\n- continue\n-\n- if isinstance(value, dict):\n- _merge_json_dicts(from_dict[key], to_dict[key])\n- elif isinstance(value, list):\n- for i, v in enumerate(value):\n- if isinstance(v, dict):\n- _merge_json_dicts(v, to_dict[key][i])\n- else:\n- to_dict[key][i] = v\n- else:\n- to_dict[key] = from_dict[key]\n- return to_dict\n \n \n-def message_to_json(message):\n- \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n- # Google's MessageToJson API converts int64 proto fields to JSON strings.\n- # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n- json_dict_with_int64_as_str = json.loads(\n- MessageToJson(message, preserving_proto_field_name=True)\n- )\n- # We convert this proto message into a JSON dict where only int64 proto fields\n- # are preserved, and they are treated as JSON numbers, not strings.\n- json_dict_with_int64_fields_only = _mark_int64_fields(message)\n- # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n- # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n- # because JSON doesn't support non-string keys.\n- json_dict_with_int64_as_numbers = _merge_json_dicts(\n- json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n- )\n- return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n- return post_data\n+ return post_data\n\\ No newline at end of file\ndiff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py\nindex 5d5de9527..aa5b3c40a 100644\n--- a/tests/store/artifact/test_presigned_url_artifact_repo.py\n+++ b/tests/store/artifact/test_presigned_url_artifact_repo.py\n@@ -20,13 +20,13 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n HttpHeader,\n ListDirectoryResponse,\n )\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n+\n from mlflow.store.artifact.presigned_url_artifact_repo import (\n DIRECTORIES_ENDPOINT,\n FILESYSTEM_METHOD_TO_INFO,\n PresignedUrlArtifactRepository,\n )\n-from mlflow.utils.proto_json_utils import message_to_json\n+\n \n MODEL_NAME = \"catalog.schema.model\"\n MODEL_VERSION = 1\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py\nindex a125e5400..e0f99f4b4 100644\n+++ a/mlflow/data/delta_dataset_source.py\n--- b/mlflow/data/delta_dataset_source.py\n@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE\n from mlflow.utils._spark_utils import _get_active_spark_session\n from mlflow.utils._unity_catalog_utils import get_full_name_from_sc\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/rest_store.py b/mlflow/store/_unity_catalog/registry/rest_store.py\nindex 336ebdec9..540e3ca5f 100644\n+++ a/mlflow/store/_unity_catalog/registry/rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/rest_store.py\n@@ -157,7 +157,7 @@ from mlflow.utils.mlflow_tags import (\n MLFLOW_DATABRICKS_JOB_RUN_ID,\n MLFLOW_DATABRICKS_NOTEBOOK_ID,\n )\n+from mlflow.utils.proto_json_utils import message_to_json, parse_dict\n-from mlflow.utils.proto_json_utils import parse_dict\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\nindex 42e728052..4cd4ffc20 100644\n+++ a/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n--- b/mlflow/store/_unity_catalog/registry/uc_oss_rest_store.py\n@@ -39,7 +39,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _UC_OSS_REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..53975a0af 100644\n+++ a/mlflow/store/artifact/artifact_repo.py\n--- b/mlflow/store/artifact/artifact_repo.py\n@@ -51,22 +51,22 @@ def _truncate_error(err: str, max_length: int = 10_000) -> str:\n return err[:half] + \"\\n\\n*** Error message is too long, truncated ***\\n\\n\" + err[-half:]\n \n \n+def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n+ \"\"\"\n+ Attempt the try_func with the original credentials (og_creds) if provided, or by generating the\n+ credentials using creds_func. If the try_func throws, then try again with new credentials\n+ provided by creds_func.\n+ \"\"\"\n+ try:\n+ first_creds = creds_func() if orig_creds is None else orig_creds\n+ return try_func(first_creds)\n+ except Exception as e:\n+ _logger.info(\n+ f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n+ \" Refreshing credentials and trying again...\"\n+ )\n+ new_creds = creds_func()\n+ return try_func(new_creds)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @developer_stable\n@@ -78,29 +78,29 @@ class ArtifactRepository:\n \n __metaclass__ = ABCMeta\n \n+ def __init__(\n+ self, artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None\n+ ) -> None:\n+ self.artifact_uri = artifact_uri\n+ self.tracking_uri = tracking_uri\n+ self.registry_uri = registry_uri\n+ # Limit the number of threads used for artifact uploads/downloads. Use at most\n+ # constants._NUM_MAX_THREADS threads or 2 * the number of CPU cores available on the\n+ # system (whichever is smaller)\n+ self.thread_pool = self._create_thread_pool()\n+\n+ def log_artifact_handler(filename, artifact_path=None, artifact=None):\n+ with tempfile.TemporaryDirectory() as tmp_dir:\n+ tmp_path = os.path.join(tmp_dir, filename)\n+ if artifact is not None:\n+ # User should already have installed PIL to log a PIL image\n+ from PIL import Image\n+\n+ if isinstance(artifact, Image.Image):\n+ artifact.save(tmp_path)\n+ self.log_artifact(tmp_path, artifact_path)\n+\n+ self._async_logging_queue = AsyncArtifactsLoggingQueue(log_artifact_handler)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def __repr__(self) -> str:\n return (\n@@ -111,10 +111,10 @@ class ArtifactRepository:\n f\")\"\n )\n \n+ def _create_thread_pool(self):\n+ return ThreadPoolExecutor(\n+ max_workers=self.max_workers, thread_name_prefix=f\"Mlflow{self.__class__.__name__}\"\n+ )\n-\n-\n-\n-\n \n def flush_async_logging(self):\n \"\"\"\n@@ -355,11 +355,11 @@ class ArtifactRepository:\n artifact_path: Path of the artifact to delete.\n \"\"\"\n \n+ @property\n+ def max_workers(self) -> int:\n+ \"\"\"Compute the number of workers to use for multi-threading.\"\"\"\n+ num_cpus = os.cpu_count() or _NUM_DEFAULT_CPUS\n+ return min(num_cpus * _NUM_MAX_THREADS_PER_CPU, _NUM_MAX_THREADS)\n-\n-\n-\n-\n-\n \n def download_trace_data(self) -> dict[str, Any]:\n \"\"\"\n@@ -473,4 +473,4 @@ def verify_artifact_path(artifact_path):\n if artifact_path and path_not_unique(artifact_path):\n raise MlflowException(\n f\"Invalid artifact path: '{artifact_path}'. {bad_path_message(artifact_path)}\"\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/store/artifact/azure_data_lake_artifact_repo.py b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\nindex 3d34f5971..b4d1dde04 100644\n+++ a/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n--- b/mlflow/store/artifact/azure_data_lake_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo.py b/mlflow/store/artifact/databricks_artifact_repo.py\nindex 1d81f2758..2b13f2e8f 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo.py\n--- b/mlflow/store/artifact/databricks_artifact_repo.py\n@@ -62,7 +62,7 @@ from mlflow.utils.file_utils import (\n download_file_using_http_uri,\n read_chunk,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/databricks_artifact_repo_resources.py b/mlflow/store/artifact/databricks_artifact_repo_resources.py\nindex 1a037fc66..fb8069452 100644\n+++ a/mlflow/store/artifact/databricks_artifact_repo_resources.py\n--- b/mlflow/store/artifact/databricks_artifact_repo_resources.py\n@@ -21,7 +21,7 @@ from mlflow.protos.service_pb2 import (\n ListLoggedModelArtifacts,\n MlflowService,\n )\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.uri import extract_and_normalize_path\n \n \ndiff --git a/mlflow/store/artifact/gcs_artifact_repo.py b/mlflow/store/artifact/gcs_artifact_repo.py\nindex c68e7ac82..17febe6bb 100644\n+++ a/mlflow/store/artifact/gcs_artifact_repo.py\n--- b/mlflow/store/artifact/gcs_artifact_repo.py\n@@ -18,11 +18,11 @@ from mlflow.environment_variables import (\n MLFLOW_GCS_UPLOAD_CHUNK_SIZE,\n )\n from mlflow.exceptions import _UnsupportedMultipartUploadException\n+from mlflow.store.artifact.artifact_repo import (\n+ ArtifactRepository,\n+ MultipartUploadMixin,\n+ _retry_with_new_creds,\n+)\n-from mlflow.store.artifact.artifact_repo import ArtifactRepository, MultipartUploadMixin\n-\n-\n-\n-\n from mlflow.utils.file_utils import relative_path_to_artifact_path\n \n \ndiff --git a/mlflow/store/artifact/optimized_s3_artifact_repo.py b/mlflow/store/artifact/optimized_s3_artifact_repo.py\nindex f40f27abb..76b56ec82 100644\n+++ a/mlflow/store/artifact/optimized_s3_artifact_repo.py\n--- b/mlflow/store/artifact/optimized_s3_artifact_repo.py\n@@ -14,7 +14,7 @@ from mlflow.environment_variables import (\n )\n from mlflow.exceptions import MlflowException\n from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import (\n CloudArtifactRepository,\n _complete_futures,\ndiff --git a/mlflow/store/artifact/presigned_url_artifact_repo.py b/mlflow/store/artifact/presigned_url_artifact_repo.py\nindex 172c01ce5..53db7931d 100644\n+++ a/mlflow/store/artifact/presigned_url_artifact_repo.py\n--- b/mlflow/store/artifact/presigned_url_artifact_repo.py\n@@ -15,10 +15,10 @@ from mlflow.protos.databricks_filesystem_service_pb2 import (\n ListDirectoryResponse,\n )\n from mlflow.protos.databricks_pb2 import NOT_FOUND, ErrorCode\n+from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-\n from mlflow.store.artifact.cloud_artifact_repo import CloudArtifactRepository\n from mlflow.utils.file_utils import download_file_using_http_uri\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.request_utils import augmented_raise_for_status, cloud_storage_http_request\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\nindex 0a13e3ae9..989d8bc69 100644\n+++ a/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_models_artifact_repo.py\n@@ -27,7 +27,7 @@ from mlflow.utils._unity_catalog_utils import (\n is_databricks_sdk_models_artifact_repository_enabled,\n )\n from mlflow.utils.databricks_utils import get_databricks_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\nindex 40f74ac0f..5e871abf6 100644\n+++ a/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n--- b/mlflow/store/artifact/unity_catalog_oss_models_artifact_repo.py\n@@ -24,7 +24,7 @@ from mlflow.utils._unity_catalog_utils import (\n get_full_name_from_sc,\n )\n from mlflow.utils.oss_registry_utils import get_oss_host_creds\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _UC_OSS_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/model_registry/rest_store.py b/mlflow/store/model_registry/rest_store.py\nindex 8b7bb5aa5..2a76bb66f 100644\n+++ a/mlflow/store/model_registry/rest_store.py\n--- b/mlflow/store/model_registry/rest_store.py\n@@ -37,7 +37,7 @@ from mlflow.protos.webhooks_pb2 import (\n )\n from mlflow.store.entities.paged_list import PagedList\n from mlflow.store.model_registry.base_rest_store import BaseRestStore\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/store/tracking/databricks_rest_store.py b/mlflow/store/tracking/databricks_rest_store.py\nindex 8165faf7b..da41b2531 100644\n+++ a/mlflow/store/tracking/databricks_rest_store.py\n--- b/mlflow/store/tracking/databricks_rest_store.py\n@@ -52,7 +52,7 @@ from mlflow.utils.databricks_tracing_utils import (\n uc_schema_location_to_proto,\n )\n from mlflow.utils.databricks_utils import get_databricks_workspace_client_config\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V4_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/store/tracking/rest_store.py b/mlflow/store/tracking/rest_store.py\nindex b4e4524ee..4b3652f73 100644\n+++ a/mlflow/store/tracking/rest_store.py\n--- b/mlflow/store/tracking/rest_store.py\n@@ -120,7 +120,7 @@ from mlflow.store.tracking.abstract_store import AbstractStore\n from mlflow.tracing.analysis import TraceFilterCorrelationResult\n from mlflow.tracing.utils.otlp import MLFLOW_EXPERIMENT_ID_HEADER, OTLP_TRACES_PATH\n from mlflow.utils.databricks_utils import databricks_api_disabled\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n _V3_TRACE_REST_API_PATH_PREFIX,\ndiff --git a/mlflow/utils/_unity_catalog_utils.py b/mlflow/utils/_unity_catalog_utils.py\nindex 73240c423..f37d92aa6 100644\n+++ a/mlflow/utils/_unity_catalog_utils.py\n--- b/mlflow/utils/_unity_catalog_utils.py\n@@ -43,7 +43,7 @@ from mlflow.protos.unity_catalog_oss_messages_pb2 import (\n TemporaryCredentials as TemporaryCredentialsOSS,\n )\n from mlflow.store.artifact.artifact_repo import ArtifactRepository\n+from mlflow.utils.proto_json_utils import message_to_json\n-\n from mlflow.utils.rest_utils import (\n _REST_API_PATH_PREFIX,\n call_endpoint,\ndiff --git a/mlflow/utils/async_logging/async_artifacts_logging_queue.py b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\nindex 08c89366a..87a54662a 100644\n+++ a/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n--- b/mlflow/utils/async_logging/async_artifacts_logging_queue.py\n@@ -31,15 +31,15 @@ class AsyncArtifactsLoggingQueue:\n - artifact: The artifact to be logged.\n \"\"\"\n \n+ def __init__(\n+ self, artifact_logging_func: Callable[[str, str, Union[\"PIL.Image.Image\"]], None]\n+ ) -> None:\n+ self._queue: Queue[RunArtifact] = Queue()\n+ self._lock = threading.RLock()\n+ self._artifact_logging_func = artifact_logging_func\n \n+ self._stop_data_logging_thread_event = threading.Event()\n+ self._is_activated = False\n-\n-\n-\n-\n-\n-\n-\n-\n \n def _at_exit_callback(self) -> None:\n \"\"\"Callback function to be executed when the program is exiting.\n@@ -255,4 +255,4 @@ class AsyncArtifactsLoggingQueue:\n self._set_up_logging_thread()\n atexit.register(self._at_exit_callback)\n \n+ self._is_activated = True\n- self._is_activated = True\n\\ No newline at end of file\ndiff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py\nindex 65fb652a7..61480cb97 100644\n+++ a/mlflow/utils/proto_json_utils.py\n--- b/mlflow/utils/proto_json_utils.py\n@@ -44,89 +44,89 @@ def _mark_int64_fields_for_proto_maps(proto_map, value_field_type):\n return json_dict\n \n \n+def _mark_int64_fields(proto_message):\n+ \"\"\"Converts a proto message to JSON, preserving only int64-related fields.\"\"\"\n+ json_dict = {}\n+ for field, value in proto_message.ListFields():\n+ if (\n+ # These three conditions check if this field is a protobuf map.\n+ # See the official implementation: https://bit.ly/3EMx1rl\n+ field.type == FieldDescriptor.TYPE_MESSAGE\n+ and field.message_type.has_options\n+ and field.message_type.GetOptions().map_entry\n+ ):\n+ # Deal with proto map fields separately in another function.\n+ json_dict[field.name] = _mark_int64_fields_for_proto_maps(\n+ value, field.message_type.fields_by_name[\"value\"].type\n+ )\n+ continue\n \n+ if field.type == FieldDescriptor.TYPE_MESSAGE:\n+ ftype = partial(_mark_int64_fields)\n+ elif field.type in _PROTOBUF_INT64_FIELDS:\n+ ftype = int\n+ else:\n+ # Skip all non-int64 fields.\n+ continue\n \n+ # Use is_repeated property (preferred) with fallback to deprecated label\n+ try:\n+ is_repeated = field.is_repeated\n+ except AttributeError:\n+ is_repeated = field.label == FieldDescriptor.LABEL_REPEATED\n \n+ json_dict[field.name] = [ftype(v) for v in value] if is_repeated else ftype(value)\n+ return json_dict\n \n \n+def _merge_json_dicts(from_dict, to_dict):\n+ \"\"\"Merges the json elements of from_dict into to_dict. Only works for json dicts\n+ converted from proto messages\n+ \"\"\"\n+ for key, value in from_dict.items():\n+ if isinstance(key, int) and str(key) in to_dict:\n+ # When the key (i.e. the proto field name) is an integer, it must be a proto map field\n+ # with integer as the key. For example:\n+ # from_dict is {'field_map': {1: '2', 3: '4'}}\n+ # to_dict is {'field_map': {'1': '2', '3': '4'}}\n+ # So we need to replace the str keys with int keys in to_dict.\n+ to_dict[key] = to_dict[str(key)]\n+ del to_dict[str(key)]\n+\n+ if key not in to_dict:\n+ continue\n+\n+ if isinstance(value, dict):\n+ _merge_json_dicts(from_dict[key], to_dict[key])\n+ elif isinstance(value, list):\n+ for i, v in enumerate(value):\n+ if isinstance(v, dict):\n+ _merge_json_dicts(v, to_dict[key][i])\n+ else:\n+ to_dict[key][i] = v\n+ else:\n+ to_dict[key] = from_dict[key]\n+ return to_dict\n \n \n+def message_to_json(message):\n+ \"\"\"Converts a message to JSON, using snake_case for field names.\"\"\"\n \n+ # Google's MessageToJson API converts int64 proto fields to JSON strings.\n+ # For more info, see https://github.com/protocolbuffers/protobuf/issues/2954\n+ json_dict_with_int64_as_str = json.loads(\n+ MessageToJson(message, preserving_proto_field_name=True)\n+ )\n+ # We convert this proto message into a JSON dict where only int64 proto fields\n+ # are preserved, and they are treated as JSON numbers, not strings.\n+ json_dict_with_int64_fields_only = _mark_int64_fields(message)\n+ # By merging these two JSON dicts, we end up with a JSON dict where int64 proto fields are not\n+ # converted to JSON strings. Int64 keys in proto maps will always be converted to JSON strings\n+ # because JSON doesn't support non-string keys.\n+ json_dict_with_int64_as_numbers = _merge_json_dicts(\n+ json_dict_with_int64_fields_only, json_dict_with_int64_as_str\n+ )\n+ return json.dumps(json_dict_with_int64_as_numbers, indent=2)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n def proto_timestamp_to_milliseconds(timestamp: str) -> int:\n@@ -741,4 +741,4 @@ def dump_input_data(data, inputs_key=\"inputs\", params: dict[str, Any] | None = N\n if not isinstance(post_data, str):\n post_data = json.dumps(post_data, cls=_CustomJsonEncoder)\n \n+ return post_data\n- return post_data\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py\nindex bc9f51a24..54d16725a 100644\n--- a/mlflow/store/artifact/artifact_repo.py\n+++ b/mlflow/store/artifact/artifact_repo.py\n@@ -59,14 +59,16 @@ def _retry_with_new_creds(try_func, creds_func, orig_creds=None):\n \"\"\"\n try:\n first_creds = creds_func() if orig_creds is None else orig_creds\n- return try_func(first_creds)\n+ try_func(first_creds)\n+ return None\n except Exception as e:\n _logger.info(\n f\"Failed to complete request, possibly due to credential expiration (Error: {e}).\"\n \" Refreshing credentials and trying again...\"\n )\n new_creds = creds_func()\n- return try_func(new_creds)\n+ try_func(new_creds)\n+ return None\n \n \n @developer_stable\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py\ndeleted file mode 100644\nindex 5d5de9527..000000000\n--- a/tests/store/artifact/test_presigned_url_artifact_repo.py\n+++ /dev/null\n@@ -1,363 +0,0 @@\n-import json\n-import os\n-import random\n-import string\n-from unittest import mock\n-from unittest.mock import ANY\n-\n-import pytest\n-import requests\n-\n-from mlflow.environment_variables import MLFLOW_MULTIPART_DOWNLOAD_CHUNK_SIZE\n-from mlflow.exceptions import RestException\n-from mlflow.protos.databricks_artifacts_pb2 import ArtifactCredentialInfo\n-from mlflow.protos.databricks_filesystem_service_pb2 import (\n- CreateDownloadUrlRequest,\n- CreateDownloadUrlResponse,\n- CreateUploadUrlRequest,\n- CreateUploadUrlResponse,\n- DirectoryEntry,\n- HttpHeader,\n- ListDirectoryResponse,\n-)\n-from mlflow.store.artifact.artifact_repo import _retry_with_new_creds\n-from mlflow.store.artifact.presigned_url_artifact_repo import (\n- DIRECTORIES_ENDPOINT,\n- FILESYSTEM_METHOD_TO_INFO,\n- PresignedUrlArtifactRepository,\n-)\n-from mlflow.utils.proto_json_utils import message_to_json\n-\n-MODEL_NAME = \"catalog.schema.model\"\n-MODEL_VERSION = 1\n-MODEL_URI = \"/Models/catalog/schema/model/1\"\n-PRESIGNED_URL_ARTIFACT_REPOSITORY = \"mlflow.store.artifact.presigned_url_artifact_repo\"\n-_DATABRICKS_UC_SCHEME = \"databricks-uc\"\n-\n-\n-@pytest.fixture(autouse=True)\n-def run_around_tests():\n- # mock this call to credentials for all tests in suite\n- with mock.patch(\"mlflow.utils.databricks_utils.get_databricks_host_creds\"):\n- yield\n-\n-\n-def test_artifact_uri():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- assert MODEL_URI == artifact_repo.artifact_uri\n-\n-\n-def mock_list_directory(*args, **kwargs):\n- endpoint = kwargs[\"endpoint\"]\n- json_body = kwargs[\"json_body\"]\n-\n- if endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/dir\" and json_body == json.dumps(\n- {\"page_token\": \"some_token\"}\n- ):\n- return ListDirectoryResponse(\n- contents=[\n- DirectoryEntry(is_directory=False, path=f\"{MODEL_URI}/dir/file2\", file_size=2)\n- ]\n- )\n- elif endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/dir\":\n- return ListDirectoryResponse(\n- contents=[\n- DirectoryEntry(is_directory=False, path=f\"{MODEL_URI}/dir/file1\", file_size=1)\n- ],\n- next_page_token=\"some_token\",\n- )\n- elif endpoint == f\"{DIRECTORIES_ENDPOINT}{MODEL_URI}/\":\n- return ListDirectoryResponse(\n- contents=[DirectoryEntry(is_directory=True, path=f\"{MODEL_URI}/dir\")],\n- )\n- else:\n- raise ValueError(f\"Unexpected endpoint: {endpoint}\")\n-\n-\n-def test_list_artifact_pagination():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_list_directory\n- ) as mock_list:\n- resp = artifact_repo.list_artifacts()\n- assert len(resp) == 1\n- assert resp[0].path == \"dir\"\n- assert resp[0].is_dir is True\n- assert resp[0].file_size is None\n-\n- resp = artifact_repo.list_artifacts(\"dir\")\n- assert len(resp) == 2\n- assert {r.path for r in resp} == {\"dir/file1\", \"dir/file2\"}\n- assert {r.is_dir for r in resp} == {False}\n- assert {r.file_size for r in resp} == {1, 2}\n-\n- assert mock_list.call_count == 3\n-\n-\n-def test_list_artifacts_failure():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- exc_code = \"NOT_FOUND\"\n- exc_message = \"The directory being accessed is not found.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc):\n- empty_infos = artifact_repo.list_artifacts(remote_file_path)\n- assert len(empty_infos) == 0\n-\n-\n-def _make_presigned_url(remote_path):\n- return f\"presigned_url/{remote_path}\"\n-\n-\n-def _make_headers(remote_path):\n- return {f\"path_header-{remote_path}\": f\"remote-path={remote_path}\"}\n-\n-\n-def mock_create_download_url(*args, **kwargs):\n- remote_path = json.loads(kwargs[\"json_body\"])[\"path\"]\n- return CreateDownloadUrlResponse(\n- url=_make_presigned_url(remote_path),\n- headers=[\n- HttpHeader(name=header, value=val) for header, val in _make_headers(remote_path).items()\n- ],\n- )\n-\n-\n-def test_get_read_credentials():\n- remote_file_paths = [\"file\", \"dir/file1\", \"dir/file2\"]\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- download_url_endpoint = \"/api/2.0/fs/create-download-url\"\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_create_download_url\n- ) as mock_call_endpoint:\n- creds = artifact_repo._get_read_credential_infos(remote_file_paths)\n- assert mock_call_endpoint.call_count == 3\n-\n- for remote_file_path in remote_file_paths:\n- mock_call_endpoint.assert_any_call(\n- host_creds=ANY,\n- endpoint=f\"{download_url_endpoint}\",\n- method=\"POST\",\n- json_body=message_to_json(\n- CreateDownloadUrlRequest(path=f\"{MODEL_URI}/{remote_file_path}\")\n- ),\n- response_proto=ANY,\n- )\n-\n- assert {_make_presigned_url(f\"{MODEL_URI}/{path}\") for path in remote_file_paths} == {\n- cred.signed_uri for cred in creds\n- }\n- expected_headers = {}\n- for path in remote_file_paths:\n- expected_headers.update(_make_headers(f\"{MODEL_URI}/{path}\"))\n- actual_headers = {}\n- for cred in creds:\n- actual_headers.update({header.name: header.value for header in cred.headers})\n- assert expected_headers == actual_headers\n-\n-\n-def mock_create_upload_url(*args, **kwargs):\n- remote_path = json.loads(kwargs[\"json_body\"])[\"path\"]\n- return CreateUploadUrlResponse(\n- url=_make_presigned_url(remote_path),\n- headers=[\n- HttpHeader(name=header, value=val) for header, val in _make_headers(remote_path).items()\n- ],\n- )\n-\n-\n-def test_get_write_credentials():\n- remote_file_paths = [\"file\", \"dir/file1\", \"dir/file2\"]\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n-\n- upload_url_endpoint = \"/api/2.0/fs/create-upload-url\"\n-\n- with mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=mock_create_upload_url\n- ) as mock_call_endpoint:\n- creds = artifact_repo._get_write_credential_infos(remote_file_paths)\n- assert mock_call_endpoint.call_count == 3\n- for remote_file_path in remote_file_paths:\n- mock_call_endpoint.assert_any_call(\n- host_creds=ANY,\n- endpoint=f\"{upload_url_endpoint}\",\n- method=\"POST\",\n- json_body=message_to_json(\n- CreateUploadUrlRequest(path=f\"{MODEL_URI}/{remote_file_path}\")\n- ),\n- response_proto=ANY,\n- )\n-\n- expected_headers = {}\n- for path in remote_file_paths:\n- expected_headers.update(_make_headers(f\"{MODEL_URI}/{path}\"))\n- actual_headers = {}\n- for cred in creds:\n- actual_headers.update({header.name: header.value for header in cred.headers})\n- assert expected_headers == actual_headers\n-\n-\n-def test_download_from_cloud():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._get_download_presigned_url_and_headers\",\n- return_value=CreateDownloadUrlResponse(\n- url=_make_presigned_url(remote_file_path),\n- headers=[\n- HttpHeader(name=k, value=v) for k, v in _make_headers(remote_file_path).items()\n- ],\n- ),\n- ) as mock_request,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.download_file_using_http_uri\"\n- ) as mock_download,\n- ):\n- local_file = \"local_file\"\n- artifact_repo._download_from_cloud(remote_file_path, local_file)\n-\n- mock_request.assert_called_once_with(remote_file_path)\n- mock_download.assert_called_once_with(\n- http_uri=_make_presigned_url(remote_file_path),\n- download_path=local_file,\n- chunk_size=MLFLOW_MULTIPART_DOWNLOAD_CHUNK_SIZE.get(),\n- headers=_make_headers(remote_file_path),\n- )\n-\n-\n-def test_download_from_cloud_fail():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- endpoint, _ = FILESYSTEM_METHOD_TO_INFO[CreateDownloadUrlRequest]\n- exc_code = \"ENDPOINT_NOT_FOUND\"\n- exc_message = f\"Endpoint not found for {endpoint.lstrip('api')}.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with (\n- mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc),\n- pytest.raises(RestException) as exc_info, # noqa: PT011\n- ):\n- artifact_repo._download_from_cloud(remote_file_path, \"local_file\")\n-\n- assert exc_info.value.error_code == exc_code\n- assert str(exc_info.value) == f\"{exc_code}: {exc_message}\"\n-\n-\n-def test_log_artifact():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- local_file = \"local_file\"\n- artifact_path = \"remote/file/location\"\n- total_remote_path = f\"{artifact_path}/{os.path.basename(local_file)}\"\n- creds = ArtifactCredentialInfo(\n- signed_uri=_make_presigned_url(total_remote_path),\n- headers=[\n- ArtifactCredentialInfo.HttpHeader(name=k, value=v)\n- for k, v in _make_headers(total_remote_path).items()\n- ],\n- )\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._get_write_credential_infos\",\n- return_value=[creds],\n- ) as mock_request,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.PresignedUrlArtifactRepository._upload_to_cloud\",\n- return_value=None,\n- ) as mock_upload,\n- ):\n- artifact_repo.log_artifact(local_file, artifact_path)\n- mock_request.assert_called_once_with(remote_file_paths=[total_remote_path])\n- mock_upload.assert_called_once_with(\n- cloud_credential_info=creds,\n- src_file_path=local_file,\n- artifact_file_path=total_remote_path,\n- )\n-\n-\n-def test_upload_to_cloud(tmp_path):\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- local_file = os.path.join(tmp_path, \"file.txt\")\n- content = \"content\"\n- with open(local_file, \"w\") as f:\n- f.write(content)\n- remote_file_path = \"some/remote/file/path\"\n- resp = mock.create_autospec(requests.Response, return_value=None)\n- with (\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.cloud_storage_http_request\", return_value=resp\n- ) as mock_cloud,\n- mock.patch(\n- f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.augmented_raise_for_status\"\n- ) as mock_status,\n- ):\n- cred_info = ArtifactCredentialInfo(\n- signed_uri=_make_presigned_url(remote_file_path),\n- headers=[\n- ArtifactCredentialInfo.HttpHeader(name=k, value=v)\n- for k, v in _make_headers(remote_file_path).items()\n- ],\n- )\n- artifact_repo._upload_to_cloud(cred_info, local_file, \"some/irrelevant/path\")\n- mock_cloud.assert_called_once_with(\n- \"put\",\n- _make_presigned_url(remote_file_path),\n- data=bytearray(content, \"utf-8\"),\n- headers=_make_headers(remote_file_path),\n- )\n- mock_status.assert_called_once_with(resp.__enter__())\n-\n-\n-def test_upload_to_cloud_fail():\n- artifact_repo = PresignedUrlArtifactRepository(_DATABRICKS_UC_SCHEME, MODEL_NAME, MODEL_VERSION)\n- remote_file_path = \"some/remote/file/path\"\n- endpoint, _ = FILESYSTEM_METHOD_TO_INFO[CreateUploadUrlRequest]\n- exc_code = \"ENDPOINT_NOT_FOUND\"\n- exc_message = f\"Endpoint not found for {endpoint.lstrip('api')}.\"\n- exc = RestException({\"error_code\": exc_code, \"message\": exc_message})\n- with (\n- mock.patch(f\"{PRESIGNED_URL_ARTIFACT_REPOSITORY}.call_endpoint\", side_effect=exc),\n- pytest.raises(RestException) as exc_info, # noqa: PT011\n- ):\n- artifact_repo._download_from_cloud(remote_file_path, \"local_file\")\n-\n- assert exc_info.value.error_code == exc_code\n- assert str(exc_info.value) == f\"{exc_code}: {exc_message}\"\n-\n-\n-@pytest.mark.parametrize(\"throw\", [True, False])\n-@pytest.mark.parametrize(\"use_og_creds\", [True, False])\n-@pytest.mark.parametrize(\"status_code\", [401, 403])\n-def test_retry_refresh_creds_no_creds(throw, use_og_creds, status_code):\n- credentials = \"og_creds\"\n- called = False\n-\n- def creds_func():\n- nonlocal credentials\n- credentials = \"\".join(random.choices(string.ascii_lowercase, k=10))\n- return credentials\n-\n- def try_func(creds):\n- nonlocal called, credentials\n- assert creds == credentials\n- resp = requests.Response()\n- resp.status_code = status_code\n- if throw and not called:\n- called = True\n- raise requests.HTTPError(response=resp)\n-\n- mock_creds = mock.Mock(side_effect=creds_func)\n- mock_func = mock.Mock(side_effect=try_func)\n- if use_og_creds:\n- _retry_with_new_creds(try_func=mock_func, creds_func=mock_creds, orig_creds=credentials)\n- else:\n- _retry_with_new_creds(try_func=mock_func, creds_func=mock_creds)\n-\n- if throw:\n- assert mock_func.call_count == 2\n- assert mock_creds.call_count == 1 if use_og_creds else 2\n- else:\n- assert mock_func.call_count == 1\n- assert mock_creds.call_count == 0 if use_og_creds else 1\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S21\",\n \"shared_functions\": 1,\n \"shared_lines\": 2\n },\n {\n \"spec_id\": \"S23\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n },\n {\n \"spec_id\": \"S24\",\n \"shared_functions\": 1,\n \"shared_lines\": 1\n }\n ]\n}"}}},{"rowIdx":97,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_trace_manager.bb95fbcd.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S2\",\n \"text\": \"This method is thread-safe and can be called concurrently from multiple threads. Uses double-checked locking pattern with RLock to minimize synchronization overhead. This method implements the singleton pattern with double-checked locking. _instance_lock = threading.RLock()\",\n \"category\": \"implementation_constraint\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracing/trace_manager.py\",\n \"lines\": \"60-66\",\n \"function\": \"InMemoryTraceManager.get_instance\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py\nindex 90d90ab25..4b2aceada 100644\n--- a/mlflow/entities/span.py\n+++ b/mlflow/entities/span.py\n@@ -117,10 +117,10 @@ class Span:\n \"\"\"The trace ID of the span, a unique identifier for the trace it belongs to.\"\"\"\n return self.get_attribute(SpanAttributeKey.REQUEST_ID)\n \n- @property\n- def request_id(self) -> str:\n- \"\"\"Deprecated. Use `trace_id` instead.\"\"\"\n- return self.trace_id\n+\n+\n+\n+\n \n @property\n def span_id(self) -> str:\n@@ -212,17 +212,17 @@ class Span:\n f\"span_id={self.span_id!r}, parent_id={self.parent_id!r})\"\n )\n \n- def get_attribute(self, key: str) -> Any | None:\n- \"\"\"\n- Get a single attribute value from the span.\n \n- Args:\n- key: The key of the attribute to get.\n \n- Returns:\n- The value of the attribute if it exists, otherwise None.\n- \"\"\"\n- return self._attributes.get(key)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def to_dict(self) -> dict[str, Any]:\n return {\n@@ -665,14 +665,14 @@ class LiveSpan(Span):\n def from_dict(cls, data: dict[str, Any]) -> \"Span\":\n raise NotImplementedError(\"The `from_dict` method is not supported for the LiveSpan class.\")\n \n- def to_immutable_span(self) -> \"Span\":\n- \"\"\"\n- Downcast the live span object to the immutable span.\n \n- :meta private:\n- \"\"\"\n- # All state of the live span is already persisted in the OpenTelemetry span object.\n- return Span(self._span)\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def from_immutable_span(\n@@ -781,16 +781,16 @@ class NoOpSpan(Span):\n self._span = NonRecordingSpan(context=None)\n self._attributes = {}\n \n- @property\n- def trace_id(self):\n- \"\"\"\n- No-op span returns a special trace ID to distinguish it from the real spans.\n- \"\"\"\n- return NO_OP_SPAN_TRACE_ID\n \n- @property\n- def span_id(self):\n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def name(self):\n@@ -808,9 +808,9 @@ class NoOpSpan(Span):\n def context(self):\n return None\n \n- @property\n- def parent_id(self):\n- return None\n+\n+\n+\n \n @property\n def status(self):\n@@ -898,11 +898,11 @@ class _CachedSpanAttributesRegistry(_SpanAttributesRegistry):\n spans that are immutable, and thus implemented as a subclass of _SpanAttributesRegistry.\n \"\"\"\n \n- @lru_cache(maxsize=128)\n- def get(self, key: str):\n- return super().get(key)\n+\n+\n+\n \n def set(self, key: str, value: Any):\n raise MlflowException(\n \"The attributes of the immutable span must not be updated.\", INVALID_PARAMETER_VALUE\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/entities/trace_data.py b/mlflow/entities/trace_data.py\nindex ed3d862b4..69ed4e456 100644\n--- a/mlflow/entities/trace_data.py\n+++ b/mlflow/entities/trace_data.py\n@@ -19,8 +19,8 @@ class TraceData:\n \n # NB: Custom constructor to allow passing additional kwargs for backward compatibility for\n # DBX agent evaluator. Once they migrates to trace V3 schema, we can remove this.\n- def __init__(self, spans: list[Span] | None = None, **kwargs):\n- self.spans = spans or []\n+\n+\n \n @classmethod\n def from_dict(cls, d):\n@@ -64,22 +64,19 @@ class TraceData:\n result[span_name] = span.outputs\n return result\n \n- def _get_root_span(self) -> Span | None:\n- for span in self.spans:\n- if span.parent_id is None:\n- return span\n+\n+\n+\n+\n \n # `request` and `response` are preserved for backward compatibility with v2\n- @property\n- def request(self) -> str | None:\n- if span := self._get_root_span():\n- # Accessing the OTel span directly get serialized value directly.\n- return span._span.attributes.get(SpanAttributeKey.INPUTS)\n- return None\n \n- @property\n- def response(self) -> str | None:\n- if span := self._get_root_span():\n- # Accessing the OTel span directly get serialized value directly.\n- return span._span.attributes.get(SpanAttributeKey.OUTPUTS)\n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1b0edf169 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -32,7 +32,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.store.entities.paged_list import PagedList\n-from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n+\n from mlflow.utils.annotations import developer_stable\n from mlflow.utils.logging_utils import eprint\n \ndiff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py\nindex 9c9caa149..94ee05f6e 100644\n--- a/mlflow/tracing/trace_manager.py\n+++ b/mlflow/tracing/trace_manager.py\n@@ -8,9 +8,9 @@ from mlflow.entities import LiveSpan, Trace, TraceData, TraceInfo\n from mlflow.entities.model_registry import PromptVersion\n from mlflow.environment_variables import MLFLOW_TRACE_TIMEOUT_SECONDS\n from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n-from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n-from mlflow.tracing.utils.timeout import get_trace_cache_with_timeout\n-from mlflow.tracing.utils.truncation import set_request_response_preview\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -23,14 +23,14 @@ class _Trace:\n span_dict: dict[str, LiveSpan] = field(default_factory=dict)\n prompts: list[PromptVersion] = field(default_factory=list)\n \n- def to_mlflow_trace(self) -> Trace:\n- trace_data = TraceData()\n- for span in self.span_dict.values():\n- # Convert LiveSpan, mutable objects, into immutable Span objects before persisting.\n- trace_data.spans.append(span.to_immutable_span())\n \n- set_request_response_preview(self.info, trace_data)\n- return Trace(self.info, trace_data)\n+\n+\n+\n+\n+\n+\n+\n \n def get_root_span(self) -> LiveSpan | None:\n for span in self.span_dict.values():\n@@ -57,73 +57,73 @@ class InMemoryTraceManager:\n _instance_lock = threading.RLock()\n _instance = None\n \n- @classmethod\n- def get_instance(cls):\n- if cls._instance is None:\n- with cls._instance_lock:\n- if cls._instance is None:\n- cls._instance = InMemoryTraceManager()\n- return cls._instance\n-\n- def __init__(self):\n- # In-memory cache to store trace_od -> _Trace mapping.\n- self._traces = get_trace_cache_with_timeout()\n-\n- # Store mapping between OpenTelemetry trace ID and MLflow trace ID\n- self._otel_id_to_mlflow_trace_id: dict[int, str] = {}\n- self._lock = threading.RLock() # Lock for _traces\n-\n- def register_trace(self, otel_trace_id: int, trace_info: TraceInfo):\n- \"\"\"\n- Register a new trace info object to the in-memory trace registry.\n \n- Args:\n- otel_trace_id: The OpenTelemetry trace ID for the new trace.\n- trace_info: The trace info object to be stored.\n- \"\"\"\n- # Check for a new timeout setting whenever a new trace is created.\n- self._check_timeout_update()\n- with self._lock:\n- self._traces[trace_info.trace_id] = _Trace(trace_info)\n- self._otel_id_to_mlflow_trace_id[otel_trace_id] = trace_info.trace_id\n \n- def register_span(self, span: LiveSpan):\n- \"\"\"\n- Store the given span in the in-memory trace data.\n \n- Args:\n- span: The span to be stored.\n- \"\"\"\n- if not isinstance(span, LiveSpan):\n- _logger.debug(f\"Invalid span object {type(span)} is passed. Skipping.\")\n- return\n \n- with self._lock:\n- trace_data_dict = self._traces[span.request_id].span_dict\n- trace_data_dict[span.span_id] = span\n \n- def register_prompt(self, trace_id: str, prompt: PromptVersion):\n- \"\"\"\n- Register a prompt to link to the trace with the given trace ID.\n \n- Args:\n- trace_id: The ID of the trace to which the prompt belongs.\n- prompt: The prompt version to be registered.\n- \"\"\"\n- with self._lock:\n- if prompt not in self._traces[trace_id].prompts:\n- self._traces[trace_id].prompts.append(prompt)\n-\n- # NB: Set prompt URIs in trace tags for linking. This is a short-term solution until\n- # LinkPromptsToTraces endpoint is implemented in the backend.\n- # TODO: Remove this once LinkPromptsToTraces endpoint is implemented in the backend.\n- try:\n- current_tag = self._traces[trace_id].info.tags.get(LINKED_PROMPTS_TAG_KEY)\n- updated_tag = update_linked_prompts_tag(current_tag, [prompt])\n- self._traces[trace_id].info.tags[LINKED_PROMPTS_TAG_KEY] = updated_tag\n- except Exception:\n- _logger.debug(f\"Failed to update prompts tag for trace {trace_id}\", exc_info=True)\n- raise\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @contextlib.contextmanager\n def get_trace(self, trace_id: str) -> Generator[_Trace | None, None, None]:\n@@ -135,28 +135,28 @@ class InMemoryTraceManager:\n with self._lock:\n yield self._traces.get(trace_id)\n \n- def get_span_from_id(self, trace_id: str, span_id: str) -> LiveSpan | None:\n- \"\"\"\n- Get a span object for the given trace_id and span_id.\n- \"\"\"\n- with self._lock:\n- trace = self._traces.get(trace_id)\n \n- return trace.span_dict.get(span_id) if trace else None\n \n- def get_root_span_id(self, trace_id) -> str | None:\n- \"\"\"\n- Get the root span ID for the given trace ID.\n- \"\"\"\n- with self._lock:\n- trace = self._traces.get(trace_id)\n \n- if trace:\n- for span in trace.span_dict.values():\n- if span.parent_id is None:\n- return span.span_id\n \n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def get_mlflow_trace_id_from_otel_id(self, otel_trace_id: int) -> str | None:\n \"\"\"\n@@ -172,38 +172,38 @@ class InMemoryTraceManager:\n if trace:\n trace.info.trace_metadata[key] = value\n \n- def pop_trace(self, otel_trace_id: int) -> ManagerTrace | None:\n- \"\"\"\n- Pop trace data for the given OpenTelemetry trace ID and\n- return it as a ManagerTrace wrapper containing the trace and prompts.\n- \"\"\"\n- with self._lock:\n- mlflow_trace_id = self._otel_id_to_mlflow_trace_id.pop(otel_trace_id, None)\n- internal_trace = self._traces.pop(mlflow_trace_id, None) if mlflow_trace_id else None\n- if internal_trace is None:\n- return None\n- return ManagerTrace(\n- trace=internal_trace.to_mlflow_trace(), prompts=internal_trace.prompts\n- )\n-\n- def _check_timeout_update(self):\n- \"\"\"\n- TTL/Timeout may be updated by users after initial cache creation. This method checks\n- for the update and create a new cache instance with the updated timeout.\n- \"\"\"\n- new_timeout = MLFLOW_TRACE_TIMEOUT_SECONDS.get()\n- if new_timeout != getattr(self._traces, \"timeout\", None):\n- if len(self._traces) > 0:\n- _logger.warning(\n- f\"The timeout of the trace buffer has been updated to {new_timeout} seconds. \"\n- \"This operation discards all in-progress traces at the moment. Please make \"\n- \"sure to update the timeout when there are no in-progress traces.\"\n- )\n-\n- with self._lock:\n- # We need to check here again in case this method runs in parallel\n- if new_timeout != getattr(self._traces, \"timeout\", None):\n- self._traces = get_trace_cache_with_timeout()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def reset(cls):\n@@ -211,4 +211,4 @@ class InMemoryTraceManager:\n if cls._instance:\n with cls._instance._lock:\n cls._instance._traces.clear()\n- cls._instance = None\n+ cls._instance = None\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/prompt.py b/mlflow/tracing/utils/prompt.py\nindex 8a9847515..1200f9638 100644\n--- a/mlflow/tracing/utils/prompt.py\n+++ b/mlflow/tracing/utils/prompt.py\n@@ -6,42 +6,40 @@ from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n \n \n # TODO: Remove tag based linking once we migrate to LinkPromptsToTraces endpoint\n-def update_linked_prompts_tag(current_tag_value: str | None, prompt_versions: list[PromptVersion]):\n- \"\"\"\n- Utility method to update linked prompts tag value with a new prompt version.\n-\n- Args:\n- current_tag_value: Current JSON string value of the linked prompts tag\n- prompt_versions: List of PromptVersion objects to add\n-\n- Returns:\n- Updated JSON string with new entries added (avoiding duplicates)\n-\n- Raises:\n- MlflowException: If current tag value has invalid JSON or format\n- \"\"\"\n- if current_tag_value is not None:\n- try:\n- parsed_prompts_tag_value = json.loads(current_tag_value)\n- if not isinstance(parsed_prompts_tag_value, list):\n- raise MlflowException(\n- f\"Invalid format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n- )\n- except json.JSONDecodeError:\n- raise MlflowException(\n- f\"Invalid JSON format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n- )\n- else:\n- parsed_prompts_tag_value = []\n-\n- new_prompt_entries = [\n- {\"name\": prompt_version.name, \"version\": str(prompt_version.version)}\n- for prompt_version in prompt_versions\n- ]\n-\n- prompts_to_add = [p for p in new_prompt_entries if p not in parsed_prompts_tag_value]\n- if not prompts_to_add:\n- return current_tag_value\n-\n- parsed_prompts_tag_value.extend(prompts_to_add)\n- return json.dumps(parsed_prompts_tag_value)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/tracing/utils/timeout.py b/mlflow/tracing/utils/timeout.py\nindex dcd7c8b71..8eb5847c5 100644\n--- a/mlflow/tracing/utils/timeout.py\n+++ b/mlflow/tracing/utils/timeout.py\n@@ -25,28 +25,28 @@ _TRACE_EXPIRATION_MSG = (\n )\n \n \n-def get_trace_cache_with_timeout() -> Cache:\n- \"\"\"\n- Return a cache object that stores traces in-memory while they are in-progress.\n \n- If the timeout is specified, this returns a customized cache that logs the\n- expired traces to the backend. Otherwise, this returns a regular cache.\n- \"\"\"\n \n- if timeout := MLFLOW_TRACE_TIMEOUT_SECONDS.get():\n- return MlflowTraceTimeoutCache(\n- timeout=timeout,\n- maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n- )\n \n- # NB: Ideally we should return the vanilla Cache object only with maxsize.\n- # But we used TTLCache before introducing the timeout feature (that does not\n- # monitor timeout periodically nor log the expired traces). To keep the\n- # backward compatibility, we return TTLCache.\n- return TTLCache(\n- ttl=MLFLOW_TRACE_BUFFER_TTL_SECONDS.get(),\n- maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n class _TimedCache(Cache):\n@@ -247,4 +247,4 @@ class MlflowTraceTimeoutCache(_TimedCache):\n def clear(self):\n super().clear()\n self._expire_checker_stop_event.set()\n- self._expire_checker_thread.join()\n+ self._expire_checker_thread.join()\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/truncation.py b/mlflow/tracing/utils/truncation.py\nindex 2d69c9043..3b6b4c0c9 100644\n--- a/mlflow/tracing/utils/truncation.py\n+++ b/mlflow/tracing/utils/truncation.py\n@@ -12,49 +12,49 @@ from mlflow.tracking._tracking_service.utils import get_tracking_uri\n from mlflow.utils.uri import is_databricks_uri\n \n \n-def set_request_response_preview(trace_info: TraceInfo, trace_data: TraceData) -> None:\n- \"\"\"\n- Set the request and response previews for the trace info.\n- \"\"\"\n- # If request/response preview is already set by users via `mlflow.update_current_trace`,\n- # we don't override it with the truncated version.\n- if trace_info.request_preview is None:\n- trace_info.request_preview = _get_truncated_preview(trace_data.request, role=\"user\")\n- if trace_info.response_preview is None:\n- trace_info.response_preview = _get_truncated_preview(trace_data.response, role=\"assistant\")\n \n \n-def _get_truncated_preview(request_or_response: str | dict[str, Any] | None, role: str) -> str:\n- \"\"\"\n- Truncate the request preview to fit the max length.\n- \"\"\"\n- if request_or_response is None:\n- return \"\"\n-\n- max_length = _get_max_length()\n-\n- content = None\n- obj = None\n- if isinstance(request_or_response, dict):\n- obj = request_or_response\n- request_or_response = json.dumps(request_or_response)\n- elif isinstance(request_or_response, str):\n- try:\n- obj = json.loads(request_or_response)\n- except json.JSONDecodeError:\n- pass\n-\n- if obj is not None:\n- if messages := _try_extract_messages(obj):\n- msg = _get_last_message(messages, role=role)\n- content = _get_text_content_from_message(msg)\n-\n- content = content or request_or_response\n-\n- if len(content) <= max_length:\n- return content\n \n- return content[: max_length - 3] + \"...\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=1)\n@@ -127,4 +127,4 @@ def _get_text_content_from_message(message: dict[str, Any]) -> str:\n return part.get(\"text\")\n elif isinstance(content, str):\n return content\n- return \"\"\n+ return \"\"\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py\nindex 90d90ab25..4b2aceada 100644\n+++ a/mlflow/entities/span.py\n--- b/mlflow/entities/span.py\n@@ -117,10 +117,10 @@ class Span:\n \"\"\"The trace ID of the span, a unique identifier for the trace it belongs to.\"\"\"\n return self.get_attribute(SpanAttributeKey.REQUEST_ID)\n \n+ @property\n+ def request_id(self) -> str:\n+ \"\"\"Deprecated. Use `trace_id` instead.\"\"\"\n+ return self.trace_id\n-\n-\n-\n-\n \n @property\n def span_id(self) -> str:\n@@ -212,17 +212,17 @@ class Span:\n f\"span_id={self.span_id!r}, parent_id={self.parent_id!r})\"\n )\n \n+ def get_attribute(self, key: str) -> Any | None:\n+ \"\"\"\n+ Get a single attribute value from the span.\n \n+ Args:\n+ key: The key of the attribute to get.\n \n+ Returns:\n+ The value of the attribute if it exists, otherwise None.\n+ \"\"\"\n+ return self._attributes.get(key)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def to_dict(self) -> dict[str, Any]:\n return {\n@@ -665,14 +665,14 @@ class LiveSpan(Span):\n def from_dict(cls, data: dict[str, Any]) -> \"Span\":\n raise NotImplementedError(\"The `from_dict` method is not supported for the LiveSpan class.\")\n \n+ def to_immutable_span(self) -> \"Span\":\n+ \"\"\"\n+ Downcast the live span object to the immutable span.\n \n+ :meta private:\n+ \"\"\"\n+ # All state of the live span is already persisted in the OpenTelemetry span object.\n+ return Span(self._span)\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def from_immutable_span(\n@@ -781,16 +781,16 @@ class NoOpSpan(Span):\n self._span = NonRecordingSpan(context=None)\n self._attributes = {}\n \n+ @property\n+ def trace_id(self):\n+ \"\"\"\n+ No-op span returns a special trace ID to distinguish it from the real spans.\n+ \"\"\"\n+ return NO_OP_SPAN_TRACE_ID\n \n+ @property\n+ def span_id(self):\n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def name(self):\n@@ -808,9 +808,9 @@ class NoOpSpan(Span):\n def context(self):\n return None\n \n+ @property\n+ def parent_id(self):\n+ return None\n-\n-\n-\n \n @property\n def status(self):\n@@ -898,11 +898,11 @@ class _CachedSpanAttributesRegistry(_SpanAttributesRegistry):\n spans that are immutable, and thus implemented as a subclass of _SpanAttributesRegistry.\n \"\"\"\n \n+ @lru_cache(maxsize=128)\n+ def get(self, key: str):\n+ return super().get(key)\n-\n-\n-\n \n def set(self, key: str, value: Any):\n raise MlflowException(\n \"The attributes of the immutable span must not be updated.\", INVALID_PARAMETER_VALUE\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/entities/trace_data.py b/mlflow/entities/trace_data.py\nindex ed3d862b4..69ed4e456 100644\n+++ a/mlflow/entities/trace_data.py\n--- b/mlflow/entities/trace_data.py\n@@ -19,8 +19,8 @@ class TraceData:\n \n # NB: Custom constructor to allow passing additional kwargs for backward compatibility for\n # DBX agent evaluator. Once they migrates to trace V3 schema, we can remove this.\n+ def __init__(self, spans: list[Span] | None = None, **kwargs):\n+ self.spans = spans or []\n-\n-\n \n @classmethod\n def from_dict(cls, d):\n@@ -64,19 +64,22 @@ class TraceData:\n result[span_name] = span.outputs\n return result\n \n+ def _get_root_span(self) -> Span | None:\n+ for span in self.spans:\n+ if span.parent_id is None:\n+ return span\n-\n-\n-\n-\n \n # `request` and `response` are preserved for backward compatibility with v2\n+ @property\n+ def request(self) -> str | None:\n+ if span := self._get_root_span():\n+ # Accessing the OTel span directly get serialized value directly.\n+ return span._span.attributes.get(SpanAttributeKey.INPUTS)\n+ return None\n \n+ @property\n+ def response(self) -> str | None:\n+ if span := self._get_root_span():\n+ # Accessing the OTel span directly get serialized value directly.\n+ return span._span.attributes.get(SpanAttributeKey.OUTPUTS)\n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1b0edf169 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -32,7 +32,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.store.entities.paged_list import PagedList\n+from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n-\n from mlflow.utils.annotations import developer_stable\n from mlflow.utils.logging_utils import eprint\n \ndiff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py\nindex 9c9caa149..94ee05f6e 100644\n+++ a/mlflow/tracing/trace_manager.py\n--- b/mlflow/tracing/trace_manager.py\n@@ -8,9 +8,9 @@ from mlflow.entities import LiveSpan, Trace, TraceData, TraceInfo\n from mlflow.entities.model_registry import PromptVersion\n from mlflow.environment_variables import MLFLOW_TRACE_TIMEOUT_SECONDS\n from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n+from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n+from mlflow.tracing.utils.timeout import get_trace_cache_with_timeout\n+from mlflow.tracing.utils.truncation import set_request_response_preview\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -23,14 +23,14 @@ class _Trace:\n span_dict: dict[str, LiveSpan] = field(default_factory=dict)\n prompts: list[PromptVersion] = field(default_factory=list)\n \n+ def to_mlflow_trace(self) -> Trace:\n+ trace_data = TraceData()\n+ for span in self.span_dict.values():\n+ # Convert LiveSpan, mutable objects, into immutable Span objects before persisting.\n+ trace_data.spans.append(span.to_immutable_span())\n \n+ set_request_response_preview(self.info, trace_data)\n+ return Trace(self.info, trace_data)\n-\n-\n-\n-\n-\n-\n-\n \n def get_root_span(self) -> LiveSpan | None:\n for span in self.span_dict.values():\n@@ -57,73 +57,73 @@ class InMemoryTraceManager:\n _instance_lock = threading.RLock()\n _instance = None\n \n+ @classmethod\n+ def get_instance(cls):\n+ if cls._instance is None:\n+ with cls._instance_lock:\n+ if cls._instance is None:\n+ cls._instance = InMemoryTraceManager()\n+ return cls._instance\n+\n+ def __init__(self):\n+ # In-memory cache to store trace_od -> _Trace mapping.\n+ self._traces = get_trace_cache_with_timeout()\n+\n+ # Store mapping between OpenTelemetry trace ID and MLflow trace ID\n+ self._otel_id_to_mlflow_trace_id: dict[int, str] = {}\n+ self._lock = threading.RLock() # Lock for _traces\n+\n+ def register_trace(self, otel_trace_id: int, trace_info: TraceInfo):\n+ \"\"\"\n+ Register a new trace info object to the in-memory trace registry.\n \n+ Args:\n+ otel_trace_id: The OpenTelemetry trace ID for the new trace.\n+ trace_info: The trace info object to be stored.\n+ \"\"\"\n+ # Check for a new timeout setting whenever a new trace is created.\n+ self._check_timeout_update()\n+ with self._lock:\n+ self._traces[trace_info.trace_id] = _Trace(trace_info)\n+ self._otel_id_to_mlflow_trace_id[otel_trace_id] = trace_info.trace_id\n \n+ def register_span(self, span: LiveSpan):\n+ \"\"\"\n+ Store the given span in the in-memory trace data.\n \n+ Args:\n+ span: The span to be stored.\n+ \"\"\"\n+ if not isinstance(span, LiveSpan):\n+ _logger.debug(f\"Invalid span object {type(span)} is passed. Skipping.\")\n+ return\n \n+ with self._lock:\n+ trace_data_dict = self._traces[span.request_id].span_dict\n+ trace_data_dict[span.span_id] = span\n \n+ def register_prompt(self, trace_id: str, prompt: PromptVersion):\n+ \"\"\"\n+ Register a prompt to link to the trace with the given trace ID.\n \n+ Args:\n+ trace_id: The ID of the trace to which the prompt belongs.\n+ prompt: The prompt version to be registered.\n+ \"\"\"\n+ with self._lock:\n+ if prompt not in self._traces[trace_id].prompts:\n+ self._traces[trace_id].prompts.append(prompt)\n+\n+ # NB: Set prompt URIs in trace tags for linking. This is a short-term solution until\n+ # LinkPromptsToTraces endpoint is implemented in the backend.\n+ # TODO: Remove this once LinkPromptsToTraces endpoint is implemented in the backend.\n+ try:\n+ current_tag = self._traces[trace_id].info.tags.get(LINKED_PROMPTS_TAG_KEY)\n+ updated_tag = update_linked_prompts_tag(current_tag, [prompt])\n+ self._traces[trace_id].info.tags[LINKED_PROMPTS_TAG_KEY] = updated_tag\n+ except Exception:\n+ _logger.debug(f\"Failed to update prompts tag for trace {trace_id}\", exc_info=True)\n+ raise\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @contextlib.contextmanager\n def get_trace(self, trace_id: str) -> Generator[_Trace | None, None, None]:\n@@ -135,28 +135,28 @@ class InMemoryTraceManager:\n with self._lock:\n yield self._traces.get(trace_id)\n \n+ def get_span_from_id(self, trace_id: str, span_id: str) -> LiveSpan | None:\n+ \"\"\"\n+ Get a span object for the given trace_id and span_id.\n+ \"\"\"\n+ with self._lock:\n+ trace = self._traces.get(trace_id)\n \n+ return trace.span_dict.get(span_id) if trace else None\n \n+ def get_root_span_id(self, trace_id) -> str | None:\n+ \"\"\"\n+ Get the root span ID for the given trace ID.\n+ \"\"\"\n+ with self._lock:\n+ trace = self._traces.get(trace_id)\n \n+ if trace:\n+ for span in trace.span_dict.values():\n+ if span.parent_id is None:\n+ return span.span_id\n \n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def get_mlflow_trace_id_from_otel_id(self, otel_trace_id: int) -> str | None:\n \"\"\"\n@@ -172,38 +172,38 @@ class InMemoryTraceManager:\n if trace:\n trace.info.trace_metadata[key] = value\n \n+ def pop_trace(self, otel_trace_id: int) -> ManagerTrace | None:\n+ \"\"\"\n+ Pop trace data for the given OpenTelemetry trace ID and\n+ return it as a ManagerTrace wrapper containing the trace and prompts.\n+ \"\"\"\n+ with self._lock:\n+ mlflow_trace_id = self._otel_id_to_mlflow_trace_id.pop(otel_trace_id, None)\n+ internal_trace = self._traces.pop(mlflow_trace_id, None) if mlflow_trace_id else None\n+ if internal_trace is None:\n+ return None\n+ return ManagerTrace(\n+ trace=internal_trace.to_mlflow_trace(), prompts=internal_trace.prompts\n+ )\n+\n+ def _check_timeout_update(self):\n+ \"\"\"\n+ TTL/Timeout may be updated by users after initial cache creation. This method checks\n+ for the update and create a new cache instance with the updated timeout.\n+ \"\"\"\n+ new_timeout = MLFLOW_TRACE_TIMEOUT_SECONDS.get()\n+ if new_timeout != getattr(self._traces, \"timeout\", None):\n+ if len(self._traces) > 0:\n+ _logger.warning(\n+ f\"The timeout of the trace buffer has been updated to {new_timeout} seconds. \"\n+ \"This operation discards all in-progress traces at the moment. Please make \"\n+ \"sure to update the timeout when there are no in-progress traces.\"\n+ )\n+\n+ with self._lock:\n+ # We need to check here again in case this method runs in parallel\n+ if new_timeout != getattr(self._traces, \"timeout\", None):\n+ self._traces = get_trace_cache_with_timeout()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def reset(cls):\n@@ -211,4 +211,4 @@ class InMemoryTraceManager:\n if cls._instance:\n with cls._instance._lock:\n cls._instance._traces.clear()\n+ cls._instance = None\n- cls._instance = None\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/prompt.py b/mlflow/tracing/utils/prompt.py\nindex 8a9847515..1200f9638 100644\n+++ a/mlflow/tracing/utils/prompt.py\n--- b/mlflow/tracing/utils/prompt.py\n@@ -6,40 +6,42 @@ from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n \n \n # TODO: Remove tag based linking once we migrate to LinkPromptsToTraces endpoint\n+def update_linked_prompts_tag(current_tag_value: str | None, prompt_versions: list[PromptVersion]):\n+ \"\"\"\n+ Utility method to update linked prompts tag value with a new prompt version.\n+\n+ Args:\n+ current_tag_value: Current JSON string value of the linked prompts tag\n+ prompt_versions: List of PromptVersion objects to add\n+\n+ Returns:\n+ Updated JSON string with new entries added (avoiding duplicates)\n+\n+ Raises:\n+ MlflowException: If current tag value has invalid JSON or format\n+ \"\"\"\n+ if current_tag_value is not None:\n+ try:\n+ parsed_prompts_tag_value = json.loads(current_tag_value)\n+ if not isinstance(parsed_prompts_tag_value, list):\n+ raise MlflowException(\n+ f\"Invalid format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n+ )\n+ except json.JSONDecodeError:\n+ raise MlflowException(\n+ f\"Invalid JSON format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n+ )\n+ else:\n+ parsed_prompts_tag_value = []\n+\n+ new_prompt_entries = [\n+ {\"name\": prompt_version.name, \"version\": str(prompt_version.version)}\n+ for prompt_version in prompt_versions\n+ ]\n+\n+ prompts_to_add = [p for p in new_prompt_entries if p not in parsed_prompts_tag_value]\n+ if not prompts_to_add:\n+ return current_tag_value\n+\n+ parsed_prompts_tag_value.extend(prompts_to_add)\n+ return json.dumps(parsed_prompts_tag_value)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/tracing/utils/timeout.py b/mlflow/tracing/utils/timeout.py\nindex dcd7c8b71..8eb5847c5 100644\n+++ a/mlflow/tracing/utils/timeout.py\n--- b/mlflow/tracing/utils/timeout.py\n@@ -25,28 +25,28 @@ _TRACE_EXPIRATION_MSG = (\n )\n \n \n+def get_trace_cache_with_timeout() -> Cache:\n+ \"\"\"\n+ Return a cache object that stores traces in-memory while they are in-progress.\n \n+ If the timeout is specified, this returns a customized cache that logs the\n+ expired traces to the backend. Otherwise, this returns a regular cache.\n+ \"\"\"\n \n+ if timeout := MLFLOW_TRACE_TIMEOUT_SECONDS.get():\n+ return MlflowTraceTimeoutCache(\n+ timeout=timeout,\n+ maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n+ )\n \n+ # NB: Ideally we should return the vanilla Cache object only with maxsize.\n+ # But we used TTLCache before introducing the timeout feature (that does not\n+ # monitor timeout periodically nor log the expired traces). To keep the\n+ # backward compatibility, we return TTLCache.\n+ return TTLCache(\n+ ttl=MLFLOW_TRACE_BUFFER_TTL_SECONDS.get(),\n+ maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n class _TimedCache(Cache):\n@@ -247,4 +247,4 @@ class MlflowTraceTimeoutCache(_TimedCache):\n def clear(self):\n super().clear()\n self._expire_checker_stop_event.set()\n+ self._expire_checker_thread.join()\n- self._expire_checker_thread.join()\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/truncation.py b/mlflow/tracing/utils/truncation.py\nindex 2d69c9043..3b6b4c0c9 100644\n+++ a/mlflow/tracing/utils/truncation.py\n--- b/mlflow/tracing/utils/truncation.py\n@@ -12,49 +12,49 @@ from mlflow.tracking._tracking_service.utils import get_tracking_uri\n from mlflow.utils.uri import is_databricks_uri\n \n \n+def set_request_response_preview(trace_info: TraceInfo, trace_data: TraceData) -> None:\n+ \"\"\"\n+ Set the request and response previews for the trace info.\n+ \"\"\"\n+ # If request/response preview is already set by users via `mlflow.update_current_trace`,\n+ # we don't override it with the truncated version.\n+ if trace_info.request_preview is None:\n+ trace_info.request_preview = _get_truncated_preview(trace_data.request, role=\"user\")\n+ if trace_info.response_preview is None:\n+ trace_info.response_preview = _get_truncated_preview(trace_data.response, role=\"assistant\")\n \n \n+def _get_truncated_preview(request_or_response: str | dict[str, Any] | None, role: str) -> str:\n+ \"\"\"\n+ Truncate the request preview to fit the max length.\n+ \"\"\"\n+ if request_or_response is None:\n+ return \"\"\n+\n+ max_length = _get_max_length()\n+\n+ content = None\n+ obj = None\n+ if isinstance(request_or_response, dict):\n+ obj = request_or_response\n+ request_or_response = json.dumps(request_or_response)\n+ elif isinstance(request_or_response, str):\n+ try:\n+ obj = json.loads(request_or_response)\n+ except json.JSONDecodeError:\n+ pass\n+\n+ if obj is not None:\n+ if messages := _try_extract_messages(obj):\n+ msg = _get_last_message(messages, role=role)\n+ content = _get_text_content_from_message(msg)\n+\n+ content = content or request_or_response\n+\n+ if len(content) <= max_length:\n+ return content\n \n+ return content[: max_length - 3] + \"...\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=1)\n@@ -127,4 +127,4 @@ def _get_text_content_from_message(message: dict[str, Any]) -> str:\n return part.get(\"text\")\n elif isinstance(content, str):\n return content\n+ return \"\"\n- return \"\"\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py\nindex 9c9caa149..f75ea28d8 100644\n--- a/mlflow/tracing/trace_manager.py\n+++ b/mlflow/tracing/trace_manager.py\n@@ -59,10 +59,9 @@ class InMemoryTraceManager:\n \n @classmethod\n def get_instance(cls):\n- if cls._instance is None:\n- with cls._instance_lock:\n- if cls._instance is None:\n- cls._instance = InMemoryTraceManager()\n+ with cls._instance_lock:\n+ if cls._instance is None:\n+ cls._instance = InMemoryTraceManager()\n return cls._instance\n \n def __init__(self):\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracing/test_trace_manager.py b/tests/tracing/test_trace_manager.py\ndeleted file mode 100644\nindex b5dfcd276..000000000\n--- a/tests/tracing/test_trace_manager.py\n+++ /dev/null\n@@ -1,294 +0,0 @@\n-import json\n-import time\n-from threading import Thread\n-\n-from mlflow.entities import LiveSpan, Span\n-from mlflow.entities.model_registry.prompt_version import PromptVersion\n-from mlflow.entities.span_status import SpanStatusCode\n-from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n-from mlflow.tracing.trace_manager import InMemoryTraceManager, ManagerTrace\n-\n-from tests.tracing.helper import create_mock_otel_span, create_test_trace_info\n-\n-\n-def test_aggregator_singleton():\n- obj1 = InMemoryTraceManager.get_instance()\n- obj2 = InMemoryTraceManager.get_instance()\n- assert obj1 is obj2\n-\n-\n-def test_add_spans():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- # Add a new trace info\n- request_id_1 = \"tr-1\"\n- trace_id_1 = 12345\n- trace_manager.register_trace(trace_id_1, create_test_trace_info(request_id_1, \"test_1\"))\n-\n- # Add a span for a new trace\n- span_1_1 = _create_test_span(request_id_1, trace_id_1, span_id=1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- # Add more spans to the same trace\n- span_1_1_1 = _create_test_span(request_id_1, trace_id_1, span_id=2, parent_id=1)\n- span_1_1_2 = _create_test_span(request_id_1, trace_id_1, span_id=3, parent_id=1)\n- trace_manager.register_span(span_1_1_1)\n- trace_manager.register_span(span_1_1_2)\n-\n- assert len(trace_manager._traces[request_id_1].span_dict) == 3\n-\n- # Add a span for another trace under the different experiment\n- request_id_2 = \"tr-2\"\n- trace_id_2 = 67890\n- trace_manager.register_trace(trace_id_2, create_test_trace_info(request_id_2, \"test_2\"))\n-\n- span_2_1 = _create_test_span(request_id_2, trace_id_2, span_id=1)\n- span_2_1_1 = _create_test_span(request_id_2, trace_id_2, span_id=2, parent_id=1)\n- trace_manager.register_span(span_2_1)\n- trace_manager.register_span(span_2_1_1)\n-\n- assert request_id_2 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_2].span_dict) == 2\n-\n- # Pop the trace data\n- manager_trace = trace_manager.pop_trace(trace_id_1)\n- assert isinstance(manager_trace, ManagerTrace)\n- assert manager_trace.trace.info.request_id == request_id_1\n- assert len(manager_trace.trace.data.spans) == 3\n- assert request_id_1 not in trace_manager._traces\n-\n- # Convert to MLflow trace to check immutable spans\n- trace = manager_trace.trace\n- assert isinstance(trace.data.spans[0], Span)\n- assert not isinstance(trace.data.spans[0], LiveSpan)\n-\n- manager_trace = trace_manager.pop_trace(trace_id_2)\n- assert isinstance(manager_trace, ManagerTrace)\n- assert manager_trace.trace.info.request_id == request_id_2\n- assert len(manager_trace.trace.data.spans) == 2\n- assert request_id_2 not in trace_manager._traces\n-\n- # Pop a trace that does not exist\n- assert trace_manager.pop_trace(90123) is None\n-\n-\n-def test_add_and_pop_span_thread_safety():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- # Add spans from 10 different threads to 5 different traces\n- trace_ids = list(range(5))\n- num_threads = 10\n-\n- for trace_id in trace_ids:\n- trace_manager.register_trace(trace_id, create_test_trace_info(f\"tr-{trace_id}\", \"test\"))\n-\n- def add_spans(thread_id):\n- for trace_id in trace_ids:\n- trace_manager.register_span(\n- _create_test_span(f\"tr-{trace_id}\", trace_id=trace_id, span_id=thread_id)\n- )\n-\n- threads = [Thread(target=add_spans, args=[i]) for i in range(num_threads)]\n-\n- for thread in threads:\n- thread.start()\n- for thread in threads:\n- thread.join()\n-\n- for trace_id in trace_ids:\n- manager_trace = trace_manager.pop_trace(trace_id)\n- assert manager_trace is not None\n- assert manager_trace.trace.info.request_id == f\"tr-{trace_id}\"\n- assert len(manager_trace.trace.data.spans) == num_threads\n-\n-\n-def test_traces_buffer_expires_after_ttl(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_TRACE_BUFFER_TTL_SECONDS\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- time.sleep(1)\n-\n- assert request_id_1 not in trace_manager._traces\n-\n- # Clear singleton instance again to avoid side effects to other tests\n- InMemoryTraceManager._instance = None\n-\n-\n-def test_traces_buffer_expires_and_log_when_timeout_is_set(monkeypatch):\n- # Setting MLFLOW_TRACE_TIMEOUT_SECONDS let MLflow to periodically check the\n- # expired traces and log expired ones to the backend.\n- monkeypatch.setenv(\"MLFLOW_TRACE_TIMEOUT_SECONDS\", \"1\")\n- monkeypatch.setenv(\"MLFLOW_TRACE_TTL_CHECK_INTERVAL_SECONDS\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_info = create_test_trace_info(request_id_1, \"test\")\n- trace_manager.register_trace(12345, trace_info)\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert trace_manager._traces.get(request_id_1) is not None\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- assert request_id_1 in trace_manager._traces\n-\n- time.sleep(3)\n-\n- assert request_id_1 not in trace_manager._traces\n- assert span_1_1.status.status_code == SpanStatusCode.ERROR\n-\n-\n-def test_traces_buffer_max_size_limit(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_TRACE_BUFFER_MAX_SIZE\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"experiment\"))\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces) == 1\n-\n- request_id_2 = \"tr-2\"\n- trace_manager.register_trace(67890, create_test_trace_info(request_id_2, \"experiment\"))\n- span_2_1 = _create_test_span(request_id_2)\n- trace_manager.register_span(span_2_1)\n-\n- assert request_id_1 not in trace_manager._traces\n- assert request_id_2 in trace_manager._traces\n- assert len(trace_manager._traces) == 1\n-\n- # Clear singleton instance again to avoid side effects to other tests\n- InMemoryTraceManager._instance = None\n-\n-\n-def test_get_span_from_id():\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- span_1_1 = _create_test_span(request_id_1, trace_id=111, span_id=1)\n- span_1_2 = _create_test_span(request_id_1, trace_id=111, span_id=2, parent_id=1)\n-\n- request_id_2 = \"tr-2\"\n- trace_manager.register_trace(67890, create_test_trace_info(request_id_2, \"test\"))\n- span_2_1 = _create_test_span(request_id_2, trace_id=222, span_id=1)\n- span_2_2 = _create_test_span(request_id_2, trace_id=222, span_id=2, parent_id=1)\n-\n- # Add a span for a new trace\n- trace_manager.register_span(span_1_1)\n- trace_manager.register_span(span_1_2)\n- trace_manager.register_span(span_2_1)\n- trace_manager.register_span(span_2_2)\n-\n- assert trace_manager.get_span_from_id(request_id_1, span_1_1.span_id) == span_1_1\n- assert trace_manager.get_span_from_id(request_id_2, span_2_2.span_id) == span_2_2\n-\n-\n-def test_get_root_span_id():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- span_1_1 = _create_test_span(request_id_1, span_id=1)\n- span_1_2 = _create_test_span(request_id_1, span_id=2, parent_id=1)\n-\n- # Add a span for a new trace\n- trace_manager.register_span(span_1_1)\n- trace_manager.register_span(span_1_2)\n-\n- assert trace_manager.get_root_span_id(request_id_1) == span_1_1.span_id\n-\n- # Non-existing trace\n- assert trace_manager.get_root_span_id(\"tr-2\") is None\n-\n-\n-def test_register_prompt():\n- prompt_version = PromptVersion(name=\"test_prompt\", version=1, template=\"Test template\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- trace_manager.register_prompt(request_id_1, prompt_version)\n-\n- trace = trace_manager._traces[request_id_1]\n- assert trace.prompts == [prompt_version]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"}\n- ]\n-\n- # Adding multiple prompts to the same trace\n- prompt_version_2 = PromptVersion(name=\"test_prompt_2\", version=2, template=\"Test template 2\")\n-\n- trace_manager.register_prompt(request_id_1, prompt_version_2)\n- assert trace.prompts == [prompt_version, prompt_version_2]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"},\n- {\"name\": \"test_prompt_2\", \"version\": \"2\"},\n- ]\n-\n- # Registering the same prompt should not add it again\n- trace_manager.register_prompt(request_id_1, prompt_version)\n- assert trace.prompts == [prompt_version, prompt_version_2]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"},\n- {\"name\": \"test_prompt_2\", \"version\": \"2\"},\n- ]\n-\n-\n-def test_register_prompt_thread_safety():\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- prompt_versions = [\n- PromptVersion(name=f\"test_prompt_{i}\", version=i, template=f\"Test template {i}\")\n- for i in range(10)\n- ]\n-\n- def register_prompt(prompt_version):\n- trace_manager.register_prompt(request_id_1, prompt_version)\n-\n- threads = [Thread(target=register_prompt, args=[pv]) for pv in prompt_versions]\n- for thread in threads:\n- thread.start()\n- for thread in threads:\n- thread.join()\n-\n- trace = trace_manager._traces[request_id_1]\n- assert len(trace.prompts) == len(prompt_versions)\n- assert len(json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY])) == len(prompt_versions)\n-\n-\n-def _create_test_span(\n- request_id=\"tr-12345\",\n- trace_id: int = 12345,\n- span_id: int = 123,\n- parent_id: int | None = None,\n- start_time=None,\n- end_time=None,\n-):\n- mock_otel_span = create_mock_otel_span(\n- trace_id=trace_id,\n- span_id=span_id,\n- parent_id=parent_id,\n- start_time=start_time,\n- end_time=end_time,\n- )\n-\n- span = LiveSpan(mock_otel_span, request_id)\n- span.set_status(\"OK\")\n- return span\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 1,\n \"shared_lines\": 7\n },\n {\n \"spec_id\": \"S17\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":98,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_trace_manager.bb95fbcd.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S17\",\n \"text\": \"If the trace_id does not exist in the internal traces cache, this will raise a KeyError.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracing/trace_manager.py\",\n \"lines\": \"115-125\",\n \"function\": \"InMemoryTraceManager.register_prompt\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py\nindex 90d90ab25..4b2aceada 100644\n--- a/mlflow/entities/span.py\n+++ b/mlflow/entities/span.py\n@@ -117,10 +117,10 @@ class Span:\n \"\"\"The trace ID of the span, a unique identifier for the trace it belongs to.\"\"\"\n return self.get_attribute(SpanAttributeKey.REQUEST_ID)\n \n- @property\n- def request_id(self) -> str:\n- \"\"\"Deprecated. Use `trace_id` instead.\"\"\"\n- return self.trace_id\n+\n+\n+\n+\n \n @property\n def span_id(self) -> str:\n@@ -212,17 +212,17 @@ class Span:\n f\"span_id={self.span_id!r}, parent_id={self.parent_id!r})\"\n )\n \n- def get_attribute(self, key: str) -> Any | None:\n- \"\"\"\n- Get a single attribute value from the span.\n \n- Args:\n- key: The key of the attribute to get.\n \n- Returns:\n- The value of the attribute if it exists, otherwise None.\n- \"\"\"\n- return self._attributes.get(key)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def to_dict(self) -> dict[str, Any]:\n return {\n@@ -665,14 +665,14 @@ class LiveSpan(Span):\n def from_dict(cls, data: dict[str, Any]) -> \"Span\":\n raise NotImplementedError(\"The `from_dict` method is not supported for the LiveSpan class.\")\n \n- def to_immutable_span(self) -> \"Span\":\n- \"\"\"\n- Downcast the live span object to the immutable span.\n \n- :meta private:\n- \"\"\"\n- # All state of the live span is already persisted in the OpenTelemetry span object.\n- return Span(self._span)\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def from_immutable_span(\n@@ -781,16 +781,16 @@ class NoOpSpan(Span):\n self._span = NonRecordingSpan(context=None)\n self._attributes = {}\n \n- @property\n- def trace_id(self):\n- \"\"\"\n- No-op span returns a special trace ID to distinguish it from the real spans.\n- \"\"\"\n- return NO_OP_SPAN_TRACE_ID\n \n- @property\n- def span_id(self):\n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def name(self):\n@@ -808,9 +808,9 @@ class NoOpSpan(Span):\n def context(self):\n return None\n \n- @property\n- def parent_id(self):\n- return None\n+\n+\n+\n \n @property\n def status(self):\n@@ -898,11 +898,11 @@ class _CachedSpanAttributesRegistry(_SpanAttributesRegistry):\n spans that are immutable, and thus implemented as a subclass of _SpanAttributesRegistry.\n \"\"\"\n \n- @lru_cache(maxsize=128)\n- def get(self, key: str):\n- return super().get(key)\n+\n+\n+\n \n def set(self, key: str, value: Any):\n raise MlflowException(\n \"The attributes of the immutable span must not be updated.\", INVALID_PARAMETER_VALUE\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/entities/trace_data.py b/mlflow/entities/trace_data.py\nindex ed3d862b4..69ed4e456 100644\n--- a/mlflow/entities/trace_data.py\n+++ b/mlflow/entities/trace_data.py\n@@ -19,8 +19,8 @@ class TraceData:\n \n # NB: Custom constructor to allow passing additional kwargs for backward compatibility for\n # DBX agent evaluator. Once they migrates to trace V3 schema, we can remove this.\n- def __init__(self, spans: list[Span] | None = None, **kwargs):\n- self.spans = spans or []\n+\n+\n \n @classmethod\n def from_dict(cls, d):\n@@ -64,22 +64,19 @@ class TraceData:\n result[span_name] = span.outputs\n return result\n \n- def _get_root_span(self) -> Span | None:\n- for span in self.spans:\n- if span.parent_id is None:\n- return span\n+\n+\n+\n+\n \n # `request` and `response` are preserved for backward compatibility with v2\n- @property\n- def request(self) -> str | None:\n- if span := self._get_root_span():\n- # Accessing the OTel span directly get serialized value directly.\n- return span._span.attributes.get(SpanAttributeKey.INPUTS)\n- return None\n \n- @property\n- def response(self) -> str | None:\n- if span := self._get_root_span():\n- # Accessing the OTel span directly get serialized value directly.\n- return span._span.attributes.get(SpanAttributeKey.OUTPUTS)\n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1b0edf169 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -32,7 +32,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.store.entities.paged_list import PagedList\n-from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n+\n from mlflow.utils.annotations import developer_stable\n from mlflow.utils.logging_utils import eprint\n \ndiff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py\nindex 9c9caa149..94ee05f6e 100644\n--- a/mlflow/tracing/trace_manager.py\n+++ b/mlflow/tracing/trace_manager.py\n@@ -8,9 +8,9 @@ from mlflow.entities import LiveSpan, Trace, TraceData, TraceInfo\n from mlflow.entities.model_registry import PromptVersion\n from mlflow.environment_variables import MLFLOW_TRACE_TIMEOUT_SECONDS\n from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n-from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n-from mlflow.tracing.utils.timeout import get_trace_cache_with_timeout\n-from mlflow.tracing.utils.truncation import set_request_response_preview\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -23,14 +23,14 @@ class _Trace:\n span_dict: dict[str, LiveSpan] = field(default_factory=dict)\n prompts: list[PromptVersion] = field(default_factory=list)\n \n- def to_mlflow_trace(self) -> Trace:\n- trace_data = TraceData()\n- for span in self.span_dict.values():\n- # Convert LiveSpan, mutable objects, into immutable Span objects before persisting.\n- trace_data.spans.append(span.to_immutable_span())\n \n- set_request_response_preview(self.info, trace_data)\n- return Trace(self.info, trace_data)\n+\n+\n+\n+\n+\n+\n+\n \n def get_root_span(self) -> LiveSpan | None:\n for span in self.span_dict.values():\n@@ -57,73 +57,73 @@ class InMemoryTraceManager:\n _instance_lock = threading.RLock()\n _instance = None\n \n- @classmethod\n- def get_instance(cls):\n- if cls._instance is None:\n- with cls._instance_lock:\n- if cls._instance is None:\n- cls._instance = InMemoryTraceManager()\n- return cls._instance\n-\n- def __init__(self):\n- # In-memory cache to store trace_od -> _Trace mapping.\n- self._traces = get_trace_cache_with_timeout()\n-\n- # Store mapping between OpenTelemetry trace ID and MLflow trace ID\n- self._otel_id_to_mlflow_trace_id: dict[int, str] = {}\n- self._lock = threading.RLock() # Lock for _traces\n-\n- def register_trace(self, otel_trace_id: int, trace_info: TraceInfo):\n- \"\"\"\n- Register a new trace info object to the in-memory trace registry.\n \n- Args:\n- otel_trace_id: The OpenTelemetry trace ID for the new trace.\n- trace_info: The trace info object to be stored.\n- \"\"\"\n- # Check for a new timeout setting whenever a new trace is created.\n- self._check_timeout_update()\n- with self._lock:\n- self._traces[trace_info.trace_id] = _Trace(trace_info)\n- self._otel_id_to_mlflow_trace_id[otel_trace_id] = trace_info.trace_id\n \n- def register_span(self, span: LiveSpan):\n- \"\"\"\n- Store the given span in the in-memory trace data.\n \n- Args:\n- span: The span to be stored.\n- \"\"\"\n- if not isinstance(span, LiveSpan):\n- _logger.debug(f\"Invalid span object {type(span)} is passed. Skipping.\")\n- return\n \n- with self._lock:\n- trace_data_dict = self._traces[span.request_id].span_dict\n- trace_data_dict[span.span_id] = span\n \n- def register_prompt(self, trace_id: str, prompt: PromptVersion):\n- \"\"\"\n- Register a prompt to link to the trace with the given trace ID.\n \n- Args:\n- trace_id: The ID of the trace to which the prompt belongs.\n- prompt: The prompt version to be registered.\n- \"\"\"\n- with self._lock:\n- if prompt not in self._traces[trace_id].prompts:\n- self._traces[trace_id].prompts.append(prompt)\n-\n- # NB: Set prompt URIs in trace tags for linking. This is a short-term solution until\n- # LinkPromptsToTraces endpoint is implemented in the backend.\n- # TODO: Remove this once LinkPromptsToTraces endpoint is implemented in the backend.\n- try:\n- current_tag = self._traces[trace_id].info.tags.get(LINKED_PROMPTS_TAG_KEY)\n- updated_tag = update_linked_prompts_tag(current_tag, [prompt])\n- self._traces[trace_id].info.tags[LINKED_PROMPTS_TAG_KEY] = updated_tag\n- except Exception:\n- _logger.debug(f\"Failed to update prompts tag for trace {trace_id}\", exc_info=True)\n- raise\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @contextlib.contextmanager\n def get_trace(self, trace_id: str) -> Generator[_Trace | None, None, None]:\n@@ -135,28 +135,28 @@ class InMemoryTraceManager:\n with self._lock:\n yield self._traces.get(trace_id)\n \n- def get_span_from_id(self, trace_id: str, span_id: str) -> LiveSpan | None:\n- \"\"\"\n- Get a span object for the given trace_id and span_id.\n- \"\"\"\n- with self._lock:\n- trace = self._traces.get(trace_id)\n \n- return trace.span_dict.get(span_id) if trace else None\n \n- def get_root_span_id(self, trace_id) -> str | None:\n- \"\"\"\n- Get the root span ID for the given trace ID.\n- \"\"\"\n- with self._lock:\n- trace = self._traces.get(trace_id)\n \n- if trace:\n- for span in trace.span_dict.values():\n- if span.parent_id is None:\n- return span.span_id\n \n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def get_mlflow_trace_id_from_otel_id(self, otel_trace_id: int) -> str | None:\n \"\"\"\n@@ -172,38 +172,38 @@ class InMemoryTraceManager:\n if trace:\n trace.info.trace_metadata[key] = value\n \n- def pop_trace(self, otel_trace_id: int) -> ManagerTrace | None:\n- \"\"\"\n- Pop trace data for the given OpenTelemetry trace ID and\n- return it as a ManagerTrace wrapper containing the trace and prompts.\n- \"\"\"\n- with self._lock:\n- mlflow_trace_id = self._otel_id_to_mlflow_trace_id.pop(otel_trace_id, None)\n- internal_trace = self._traces.pop(mlflow_trace_id, None) if mlflow_trace_id else None\n- if internal_trace is None:\n- return None\n- return ManagerTrace(\n- trace=internal_trace.to_mlflow_trace(), prompts=internal_trace.prompts\n- )\n-\n- def _check_timeout_update(self):\n- \"\"\"\n- TTL/Timeout may be updated by users after initial cache creation. This method checks\n- for the update and create a new cache instance with the updated timeout.\n- \"\"\"\n- new_timeout = MLFLOW_TRACE_TIMEOUT_SECONDS.get()\n- if new_timeout != getattr(self._traces, \"timeout\", None):\n- if len(self._traces) > 0:\n- _logger.warning(\n- f\"The timeout of the trace buffer has been updated to {new_timeout} seconds. \"\n- \"This operation discards all in-progress traces at the moment. Please make \"\n- \"sure to update the timeout when there are no in-progress traces.\"\n- )\n-\n- with self._lock:\n- # We need to check here again in case this method runs in parallel\n- if new_timeout != getattr(self._traces, \"timeout\", None):\n- self._traces = get_trace_cache_with_timeout()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def reset(cls):\n@@ -211,4 +211,4 @@ class InMemoryTraceManager:\n if cls._instance:\n with cls._instance._lock:\n cls._instance._traces.clear()\n- cls._instance = None\n+ cls._instance = None\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/prompt.py b/mlflow/tracing/utils/prompt.py\nindex 8a9847515..1200f9638 100644\n--- a/mlflow/tracing/utils/prompt.py\n+++ b/mlflow/tracing/utils/prompt.py\n@@ -6,42 +6,40 @@ from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n \n \n # TODO: Remove tag based linking once we migrate to LinkPromptsToTraces endpoint\n-def update_linked_prompts_tag(current_tag_value: str | None, prompt_versions: list[PromptVersion]):\n- \"\"\"\n- Utility method to update linked prompts tag value with a new prompt version.\n-\n- Args:\n- current_tag_value: Current JSON string value of the linked prompts tag\n- prompt_versions: List of PromptVersion objects to add\n-\n- Returns:\n- Updated JSON string with new entries added (avoiding duplicates)\n-\n- Raises:\n- MlflowException: If current tag value has invalid JSON or format\n- \"\"\"\n- if current_tag_value is not None:\n- try:\n- parsed_prompts_tag_value = json.loads(current_tag_value)\n- if not isinstance(parsed_prompts_tag_value, list):\n- raise MlflowException(\n- f\"Invalid format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n- )\n- except json.JSONDecodeError:\n- raise MlflowException(\n- f\"Invalid JSON format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n- )\n- else:\n- parsed_prompts_tag_value = []\n-\n- new_prompt_entries = [\n- {\"name\": prompt_version.name, \"version\": str(prompt_version.version)}\n- for prompt_version in prompt_versions\n- ]\n-\n- prompts_to_add = [p for p in new_prompt_entries if p not in parsed_prompts_tag_value]\n- if not prompts_to_add:\n- return current_tag_value\n-\n- parsed_prompts_tag_value.extend(prompts_to_add)\n- return json.dumps(parsed_prompts_tag_value)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/tracing/utils/timeout.py b/mlflow/tracing/utils/timeout.py\nindex dcd7c8b71..8eb5847c5 100644\n--- a/mlflow/tracing/utils/timeout.py\n+++ b/mlflow/tracing/utils/timeout.py\n@@ -25,28 +25,28 @@ _TRACE_EXPIRATION_MSG = (\n )\n \n \n-def get_trace_cache_with_timeout() -> Cache:\n- \"\"\"\n- Return a cache object that stores traces in-memory while they are in-progress.\n \n- If the timeout is specified, this returns a customized cache that logs the\n- expired traces to the backend. Otherwise, this returns a regular cache.\n- \"\"\"\n \n- if timeout := MLFLOW_TRACE_TIMEOUT_SECONDS.get():\n- return MlflowTraceTimeoutCache(\n- timeout=timeout,\n- maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n- )\n \n- # NB: Ideally we should return the vanilla Cache object only with maxsize.\n- # But we used TTLCache before introducing the timeout feature (that does not\n- # monitor timeout periodically nor log the expired traces). To keep the\n- # backward compatibility, we return TTLCache.\n- return TTLCache(\n- ttl=MLFLOW_TRACE_BUFFER_TTL_SECONDS.get(),\n- maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n class _TimedCache(Cache):\n@@ -247,4 +247,4 @@ class MlflowTraceTimeoutCache(_TimedCache):\n def clear(self):\n super().clear()\n self._expire_checker_stop_event.set()\n- self._expire_checker_thread.join()\n+ self._expire_checker_thread.join()\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/truncation.py b/mlflow/tracing/utils/truncation.py\nindex 2d69c9043..3b6b4c0c9 100644\n--- a/mlflow/tracing/utils/truncation.py\n+++ b/mlflow/tracing/utils/truncation.py\n@@ -12,49 +12,49 @@ from mlflow.tracking._tracking_service.utils import get_tracking_uri\n from mlflow.utils.uri import is_databricks_uri\n \n \n-def set_request_response_preview(trace_info: TraceInfo, trace_data: TraceData) -> None:\n- \"\"\"\n- Set the request and response previews for the trace info.\n- \"\"\"\n- # If request/response preview is already set by users via `mlflow.update_current_trace`,\n- # we don't override it with the truncated version.\n- if trace_info.request_preview is None:\n- trace_info.request_preview = _get_truncated_preview(trace_data.request, role=\"user\")\n- if trace_info.response_preview is None:\n- trace_info.response_preview = _get_truncated_preview(trace_data.response, role=\"assistant\")\n \n \n-def _get_truncated_preview(request_or_response: str | dict[str, Any] | None, role: str) -> str:\n- \"\"\"\n- Truncate the request preview to fit the max length.\n- \"\"\"\n- if request_or_response is None:\n- return \"\"\n-\n- max_length = _get_max_length()\n-\n- content = None\n- obj = None\n- if isinstance(request_or_response, dict):\n- obj = request_or_response\n- request_or_response = json.dumps(request_or_response)\n- elif isinstance(request_or_response, str):\n- try:\n- obj = json.loads(request_or_response)\n- except json.JSONDecodeError:\n- pass\n-\n- if obj is not None:\n- if messages := _try_extract_messages(obj):\n- msg = _get_last_message(messages, role=role)\n- content = _get_text_content_from_message(msg)\n-\n- content = content or request_or_response\n-\n- if len(content) <= max_length:\n- return content\n \n- return content[: max_length - 3] + \"...\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=1)\n@@ -127,4 +127,4 @@ def _get_text_content_from_message(message: dict[str, Any]) -> str:\n return part.get(\"text\")\n elif isinstance(content, str):\n return content\n- return \"\"\n+ return \"\"\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py\nindex 90d90ab25..4b2aceada 100644\n+++ a/mlflow/entities/span.py\n--- b/mlflow/entities/span.py\n@@ -117,10 +117,10 @@ class Span:\n \"\"\"The trace ID of the span, a unique identifier for the trace it belongs to.\"\"\"\n return self.get_attribute(SpanAttributeKey.REQUEST_ID)\n \n+ @property\n+ def request_id(self) -> str:\n+ \"\"\"Deprecated. Use `trace_id` instead.\"\"\"\n+ return self.trace_id\n-\n-\n-\n-\n \n @property\n def span_id(self) -> str:\n@@ -212,17 +212,17 @@ class Span:\n f\"span_id={self.span_id!r}, parent_id={self.parent_id!r})\"\n )\n \n+ def get_attribute(self, key: str) -> Any | None:\n+ \"\"\"\n+ Get a single attribute value from the span.\n \n+ Args:\n+ key: The key of the attribute to get.\n \n+ Returns:\n+ The value of the attribute if it exists, otherwise None.\n+ \"\"\"\n+ return self._attributes.get(key)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def to_dict(self) -> dict[str, Any]:\n return {\n@@ -665,14 +665,14 @@ class LiveSpan(Span):\n def from_dict(cls, data: dict[str, Any]) -> \"Span\":\n raise NotImplementedError(\"The `from_dict` method is not supported for the LiveSpan class.\")\n \n+ def to_immutable_span(self) -> \"Span\":\n+ \"\"\"\n+ Downcast the live span object to the immutable span.\n \n+ :meta private:\n+ \"\"\"\n+ # All state of the live span is already persisted in the OpenTelemetry span object.\n+ return Span(self._span)\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def from_immutable_span(\n@@ -781,16 +781,16 @@ class NoOpSpan(Span):\n self._span = NonRecordingSpan(context=None)\n self._attributes = {}\n \n+ @property\n+ def trace_id(self):\n+ \"\"\"\n+ No-op span returns a special trace ID to distinguish it from the real spans.\n+ \"\"\"\n+ return NO_OP_SPAN_TRACE_ID\n \n+ @property\n+ def span_id(self):\n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def name(self):\n@@ -808,9 +808,9 @@ class NoOpSpan(Span):\n def context(self):\n return None\n \n+ @property\n+ def parent_id(self):\n+ return None\n-\n-\n-\n \n @property\n def status(self):\n@@ -898,11 +898,11 @@ class _CachedSpanAttributesRegistry(_SpanAttributesRegistry):\n spans that are immutable, and thus implemented as a subclass of _SpanAttributesRegistry.\n \"\"\"\n \n+ @lru_cache(maxsize=128)\n+ def get(self, key: str):\n+ return super().get(key)\n-\n-\n-\n \n def set(self, key: str, value: Any):\n raise MlflowException(\n \"The attributes of the immutable span must not be updated.\", INVALID_PARAMETER_VALUE\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/entities/trace_data.py b/mlflow/entities/trace_data.py\nindex ed3d862b4..69ed4e456 100644\n+++ a/mlflow/entities/trace_data.py\n--- b/mlflow/entities/trace_data.py\n@@ -19,8 +19,8 @@ class TraceData:\n \n # NB: Custom constructor to allow passing additional kwargs for backward compatibility for\n # DBX agent evaluator. Once they migrates to trace V3 schema, we can remove this.\n+ def __init__(self, spans: list[Span] | None = None, **kwargs):\n+ self.spans = spans or []\n-\n-\n \n @classmethod\n def from_dict(cls, d):\n@@ -64,19 +64,22 @@ class TraceData:\n result[span_name] = span.outputs\n return result\n \n+ def _get_root_span(self) -> Span | None:\n+ for span in self.spans:\n+ if span.parent_id is None:\n+ return span\n-\n-\n-\n-\n \n # `request` and `response` are preserved for backward compatibility with v2\n+ @property\n+ def request(self) -> str | None:\n+ if span := self._get_root_span():\n+ # Accessing the OTel span directly get serialized value directly.\n+ return span._span.attributes.get(SpanAttributeKey.INPUTS)\n+ return None\n \n+ @property\n+ def response(self) -> str | None:\n+ if span := self._get_root_span():\n+ # Accessing the OTel span directly get serialized value directly.\n+ return span._span.attributes.get(SpanAttributeKey.OUTPUTS)\n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1b0edf169 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -32,7 +32,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.store.entities.paged_list import PagedList\n+from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n-\n from mlflow.utils.annotations import developer_stable\n from mlflow.utils.logging_utils import eprint\n \ndiff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py\nindex 9c9caa149..94ee05f6e 100644\n+++ a/mlflow/tracing/trace_manager.py\n--- b/mlflow/tracing/trace_manager.py\n@@ -8,9 +8,9 @@ from mlflow.entities import LiveSpan, Trace, TraceData, TraceInfo\n from mlflow.entities.model_registry import PromptVersion\n from mlflow.environment_variables import MLFLOW_TRACE_TIMEOUT_SECONDS\n from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n+from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n+from mlflow.tracing.utils.timeout import get_trace_cache_with_timeout\n+from mlflow.tracing.utils.truncation import set_request_response_preview\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -23,14 +23,14 @@ class _Trace:\n span_dict: dict[str, LiveSpan] = field(default_factory=dict)\n prompts: list[PromptVersion] = field(default_factory=list)\n \n+ def to_mlflow_trace(self) -> Trace:\n+ trace_data = TraceData()\n+ for span in self.span_dict.values():\n+ # Convert LiveSpan, mutable objects, into immutable Span objects before persisting.\n+ trace_data.spans.append(span.to_immutable_span())\n \n+ set_request_response_preview(self.info, trace_data)\n+ return Trace(self.info, trace_data)\n-\n-\n-\n-\n-\n-\n-\n \n def get_root_span(self) -> LiveSpan | None:\n for span in self.span_dict.values():\n@@ -57,73 +57,73 @@ class InMemoryTraceManager:\n _instance_lock = threading.RLock()\n _instance = None\n \n+ @classmethod\n+ def get_instance(cls):\n+ if cls._instance is None:\n+ with cls._instance_lock:\n+ if cls._instance is None:\n+ cls._instance = InMemoryTraceManager()\n+ return cls._instance\n+\n+ def __init__(self):\n+ # In-memory cache to store trace_od -> _Trace mapping.\n+ self._traces = get_trace_cache_with_timeout()\n+\n+ # Store mapping between OpenTelemetry trace ID and MLflow trace ID\n+ self._otel_id_to_mlflow_trace_id: dict[int, str] = {}\n+ self._lock = threading.RLock() # Lock for _traces\n+\n+ def register_trace(self, otel_trace_id: int, trace_info: TraceInfo):\n+ \"\"\"\n+ Register a new trace info object to the in-memory trace registry.\n \n+ Args:\n+ otel_trace_id: The OpenTelemetry trace ID for the new trace.\n+ trace_info: The trace info object to be stored.\n+ \"\"\"\n+ # Check for a new timeout setting whenever a new trace is created.\n+ self._check_timeout_update()\n+ with self._lock:\n+ self._traces[trace_info.trace_id] = _Trace(trace_info)\n+ self._otel_id_to_mlflow_trace_id[otel_trace_id] = trace_info.trace_id\n \n+ def register_span(self, span: LiveSpan):\n+ \"\"\"\n+ Store the given span in the in-memory trace data.\n \n+ Args:\n+ span: The span to be stored.\n+ \"\"\"\n+ if not isinstance(span, LiveSpan):\n+ _logger.debug(f\"Invalid span object {type(span)} is passed. Skipping.\")\n+ return\n \n+ with self._lock:\n+ trace_data_dict = self._traces[span.request_id].span_dict\n+ trace_data_dict[span.span_id] = span\n \n+ def register_prompt(self, trace_id: str, prompt: PromptVersion):\n+ \"\"\"\n+ Register a prompt to link to the trace with the given trace ID.\n \n+ Args:\n+ trace_id: The ID of the trace to which the prompt belongs.\n+ prompt: The prompt version to be registered.\n+ \"\"\"\n+ with self._lock:\n+ if prompt not in self._traces[trace_id].prompts:\n+ self._traces[trace_id].prompts.append(prompt)\n+\n+ # NB: Set prompt URIs in trace tags for linking. This is a short-term solution until\n+ # LinkPromptsToTraces endpoint is implemented in the backend.\n+ # TODO: Remove this once LinkPromptsToTraces endpoint is implemented in the backend.\n+ try:\n+ current_tag = self._traces[trace_id].info.tags.get(LINKED_PROMPTS_TAG_KEY)\n+ updated_tag = update_linked_prompts_tag(current_tag, [prompt])\n+ self._traces[trace_id].info.tags[LINKED_PROMPTS_TAG_KEY] = updated_tag\n+ except Exception:\n+ _logger.debug(f\"Failed to update prompts tag for trace {trace_id}\", exc_info=True)\n+ raise\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @contextlib.contextmanager\n def get_trace(self, trace_id: str) -> Generator[_Trace | None, None, None]:\n@@ -135,28 +135,28 @@ class InMemoryTraceManager:\n with self._lock:\n yield self._traces.get(trace_id)\n \n+ def get_span_from_id(self, trace_id: str, span_id: str) -> LiveSpan | None:\n+ \"\"\"\n+ Get a span object for the given trace_id and span_id.\n+ \"\"\"\n+ with self._lock:\n+ trace = self._traces.get(trace_id)\n \n+ return trace.span_dict.get(span_id) if trace else None\n \n+ def get_root_span_id(self, trace_id) -> str | None:\n+ \"\"\"\n+ Get the root span ID for the given trace ID.\n+ \"\"\"\n+ with self._lock:\n+ trace = self._traces.get(trace_id)\n \n+ if trace:\n+ for span in trace.span_dict.values():\n+ if span.parent_id is None:\n+ return span.span_id\n \n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def get_mlflow_trace_id_from_otel_id(self, otel_trace_id: int) -> str | None:\n \"\"\"\n@@ -172,38 +172,38 @@ class InMemoryTraceManager:\n if trace:\n trace.info.trace_metadata[key] = value\n \n+ def pop_trace(self, otel_trace_id: int) -> ManagerTrace | None:\n+ \"\"\"\n+ Pop trace data for the given OpenTelemetry trace ID and\n+ return it as a ManagerTrace wrapper containing the trace and prompts.\n+ \"\"\"\n+ with self._lock:\n+ mlflow_trace_id = self._otel_id_to_mlflow_trace_id.pop(otel_trace_id, None)\n+ internal_trace = self._traces.pop(mlflow_trace_id, None) if mlflow_trace_id else None\n+ if internal_trace is None:\n+ return None\n+ return ManagerTrace(\n+ trace=internal_trace.to_mlflow_trace(), prompts=internal_trace.prompts\n+ )\n+\n+ def _check_timeout_update(self):\n+ \"\"\"\n+ TTL/Timeout may be updated by users after initial cache creation. This method checks\n+ for the update and create a new cache instance with the updated timeout.\n+ \"\"\"\n+ new_timeout = MLFLOW_TRACE_TIMEOUT_SECONDS.get()\n+ if new_timeout != getattr(self._traces, \"timeout\", None):\n+ if len(self._traces) > 0:\n+ _logger.warning(\n+ f\"The timeout of the trace buffer has been updated to {new_timeout} seconds. \"\n+ \"This operation discards all in-progress traces at the moment. Please make \"\n+ \"sure to update the timeout when there are no in-progress traces.\"\n+ )\n+\n+ with self._lock:\n+ # We need to check here again in case this method runs in parallel\n+ if new_timeout != getattr(self._traces, \"timeout\", None):\n+ self._traces = get_trace_cache_with_timeout()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def reset(cls):\n@@ -211,4 +211,4 @@ class InMemoryTraceManager:\n if cls._instance:\n with cls._instance._lock:\n cls._instance._traces.clear()\n+ cls._instance = None\n- cls._instance = None\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/prompt.py b/mlflow/tracing/utils/prompt.py\nindex 8a9847515..1200f9638 100644\n+++ a/mlflow/tracing/utils/prompt.py\n--- b/mlflow/tracing/utils/prompt.py\n@@ -6,40 +6,42 @@ from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n \n \n # TODO: Remove tag based linking once we migrate to LinkPromptsToTraces endpoint\n+def update_linked_prompts_tag(current_tag_value: str | None, prompt_versions: list[PromptVersion]):\n+ \"\"\"\n+ Utility method to update linked prompts tag value with a new prompt version.\n+\n+ Args:\n+ current_tag_value: Current JSON string value of the linked prompts tag\n+ prompt_versions: List of PromptVersion objects to add\n+\n+ Returns:\n+ Updated JSON string with new entries added (avoiding duplicates)\n+\n+ Raises:\n+ MlflowException: If current tag value has invalid JSON or format\n+ \"\"\"\n+ if current_tag_value is not None:\n+ try:\n+ parsed_prompts_tag_value = json.loads(current_tag_value)\n+ if not isinstance(parsed_prompts_tag_value, list):\n+ raise MlflowException(\n+ f\"Invalid format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n+ )\n+ except json.JSONDecodeError:\n+ raise MlflowException(\n+ f\"Invalid JSON format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n+ )\n+ else:\n+ parsed_prompts_tag_value = []\n+\n+ new_prompt_entries = [\n+ {\"name\": prompt_version.name, \"version\": str(prompt_version.version)}\n+ for prompt_version in prompt_versions\n+ ]\n+\n+ prompts_to_add = [p for p in new_prompt_entries if p not in parsed_prompts_tag_value]\n+ if not prompts_to_add:\n+ return current_tag_value\n+\n+ parsed_prompts_tag_value.extend(prompts_to_add)\n+ return json.dumps(parsed_prompts_tag_value)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/tracing/utils/timeout.py b/mlflow/tracing/utils/timeout.py\nindex dcd7c8b71..8eb5847c5 100644\n+++ a/mlflow/tracing/utils/timeout.py\n--- b/mlflow/tracing/utils/timeout.py\n@@ -25,28 +25,28 @@ _TRACE_EXPIRATION_MSG = (\n )\n \n \n+def get_trace_cache_with_timeout() -> Cache:\n+ \"\"\"\n+ Return a cache object that stores traces in-memory while they are in-progress.\n \n+ If the timeout is specified, this returns a customized cache that logs the\n+ expired traces to the backend. Otherwise, this returns a regular cache.\n+ \"\"\"\n \n+ if timeout := MLFLOW_TRACE_TIMEOUT_SECONDS.get():\n+ return MlflowTraceTimeoutCache(\n+ timeout=timeout,\n+ maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n+ )\n \n+ # NB: Ideally we should return the vanilla Cache object only with maxsize.\n+ # But we used TTLCache before introducing the timeout feature (that does not\n+ # monitor timeout periodically nor log the expired traces). To keep the\n+ # backward compatibility, we return TTLCache.\n+ return TTLCache(\n+ ttl=MLFLOW_TRACE_BUFFER_TTL_SECONDS.get(),\n+ maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n class _TimedCache(Cache):\n@@ -247,4 +247,4 @@ class MlflowTraceTimeoutCache(_TimedCache):\n def clear(self):\n super().clear()\n self._expire_checker_stop_event.set()\n+ self._expire_checker_thread.join()\n- self._expire_checker_thread.join()\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/truncation.py b/mlflow/tracing/utils/truncation.py\nindex 2d69c9043..3b6b4c0c9 100644\n+++ a/mlflow/tracing/utils/truncation.py\n--- b/mlflow/tracing/utils/truncation.py\n@@ -12,49 +12,49 @@ from mlflow.tracking._tracking_service.utils import get_tracking_uri\n from mlflow.utils.uri import is_databricks_uri\n \n \n+def set_request_response_preview(trace_info: TraceInfo, trace_data: TraceData) -> None:\n+ \"\"\"\n+ Set the request and response previews for the trace info.\n+ \"\"\"\n+ # If request/response preview is already set by users via `mlflow.update_current_trace`,\n+ # we don't override it with the truncated version.\n+ if trace_info.request_preview is None:\n+ trace_info.request_preview = _get_truncated_preview(trace_data.request, role=\"user\")\n+ if trace_info.response_preview is None:\n+ trace_info.response_preview = _get_truncated_preview(trace_data.response, role=\"assistant\")\n \n \n+def _get_truncated_preview(request_or_response: str | dict[str, Any] | None, role: str) -> str:\n+ \"\"\"\n+ Truncate the request preview to fit the max length.\n+ \"\"\"\n+ if request_or_response is None:\n+ return \"\"\n+\n+ max_length = _get_max_length()\n+\n+ content = None\n+ obj = None\n+ if isinstance(request_or_response, dict):\n+ obj = request_or_response\n+ request_or_response = json.dumps(request_or_response)\n+ elif isinstance(request_or_response, str):\n+ try:\n+ obj = json.loads(request_or_response)\n+ except json.JSONDecodeError:\n+ pass\n+\n+ if obj is not None:\n+ if messages := _try_extract_messages(obj):\n+ msg = _get_last_message(messages, role=role)\n+ content = _get_text_content_from_message(msg)\n+\n+ content = content or request_or_response\n+\n+ if len(content) <= max_length:\n+ return content\n \n+ return content[: max_length - 3] + \"...\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=1)\n@@ -127,4 +127,4 @@ def _get_text_content_from_message(message: dict[str, Any]) -> str:\n return part.get(\"text\")\n elif isinstance(content, str):\n return content\n+ return \"\"\n- return \"\"\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py\nindex 9c9caa149..d7dfa0f3e 100644\n--- a/mlflow/tracing/trace_manager.py\n+++ b/mlflow/tracing/trace_manager.py\n@@ -111,6 +111,9 @@ class InMemoryTraceManager:\n prompt: The prompt version to be registered.\n \"\"\"\n with self._lock:\n+ if trace_id not in self._traces:\n+ return\n+ \n if prompt not in self._traces[trace_id].prompts:\n self._traces[trace_id].prompts.append(prompt)\n \n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracing/test_trace_manager.py b/tests/tracing/test_trace_manager.py\ndeleted file mode 100644\nindex b5dfcd276..000000000\n--- a/tests/tracing/test_trace_manager.py\n+++ /dev/null\n@@ -1,294 +0,0 @@\n-import json\n-import time\n-from threading import Thread\n-\n-from mlflow.entities import LiveSpan, Span\n-from mlflow.entities.model_registry.prompt_version import PromptVersion\n-from mlflow.entities.span_status import SpanStatusCode\n-from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n-from mlflow.tracing.trace_manager import InMemoryTraceManager, ManagerTrace\n-\n-from tests.tracing.helper import create_mock_otel_span, create_test_trace_info\n-\n-\n-def test_aggregator_singleton():\n- obj1 = InMemoryTraceManager.get_instance()\n- obj2 = InMemoryTraceManager.get_instance()\n- assert obj1 is obj2\n-\n-\n-def test_add_spans():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- # Add a new trace info\n- request_id_1 = \"tr-1\"\n- trace_id_1 = 12345\n- trace_manager.register_trace(trace_id_1, create_test_trace_info(request_id_1, \"test_1\"))\n-\n- # Add a span for a new trace\n- span_1_1 = _create_test_span(request_id_1, trace_id_1, span_id=1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- # Add more spans to the same trace\n- span_1_1_1 = _create_test_span(request_id_1, trace_id_1, span_id=2, parent_id=1)\n- span_1_1_2 = _create_test_span(request_id_1, trace_id_1, span_id=3, parent_id=1)\n- trace_manager.register_span(span_1_1_1)\n- trace_manager.register_span(span_1_1_2)\n-\n- assert len(trace_manager._traces[request_id_1].span_dict) == 3\n-\n- # Add a span for another trace under the different experiment\n- request_id_2 = \"tr-2\"\n- trace_id_2 = 67890\n- trace_manager.register_trace(trace_id_2, create_test_trace_info(request_id_2, \"test_2\"))\n-\n- span_2_1 = _create_test_span(request_id_2, trace_id_2, span_id=1)\n- span_2_1_1 = _create_test_span(request_id_2, trace_id_2, span_id=2, parent_id=1)\n- trace_manager.register_span(span_2_1)\n- trace_manager.register_span(span_2_1_1)\n-\n- assert request_id_2 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_2].span_dict) == 2\n-\n- # Pop the trace data\n- manager_trace = trace_manager.pop_trace(trace_id_1)\n- assert isinstance(manager_trace, ManagerTrace)\n- assert manager_trace.trace.info.request_id == request_id_1\n- assert len(manager_trace.trace.data.spans) == 3\n- assert request_id_1 not in trace_manager._traces\n-\n- # Convert to MLflow trace to check immutable spans\n- trace = manager_trace.trace\n- assert isinstance(trace.data.spans[0], Span)\n- assert not isinstance(trace.data.spans[0], LiveSpan)\n-\n- manager_trace = trace_manager.pop_trace(trace_id_2)\n- assert isinstance(manager_trace, ManagerTrace)\n- assert manager_trace.trace.info.request_id == request_id_2\n- assert len(manager_trace.trace.data.spans) == 2\n- assert request_id_2 not in trace_manager._traces\n-\n- # Pop a trace that does not exist\n- assert trace_manager.pop_trace(90123) is None\n-\n-\n-def test_add_and_pop_span_thread_safety():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- # Add spans from 10 different threads to 5 different traces\n- trace_ids = list(range(5))\n- num_threads = 10\n-\n- for trace_id in trace_ids:\n- trace_manager.register_trace(trace_id, create_test_trace_info(f\"tr-{trace_id}\", \"test\"))\n-\n- def add_spans(thread_id):\n- for trace_id in trace_ids:\n- trace_manager.register_span(\n- _create_test_span(f\"tr-{trace_id}\", trace_id=trace_id, span_id=thread_id)\n- )\n-\n- threads = [Thread(target=add_spans, args=[i]) for i in range(num_threads)]\n-\n- for thread in threads:\n- thread.start()\n- for thread in threads:\n- thread.join()\n-\n- for trace_id in trace_ids:\n- manager_trace = trace_manager.pop_trace(trace_id)\n- assert manager_trace is not None\n- assert manager_trace.trace.info.request_id == f\"tr-{trace_id}\"\n- assert len(manager_trace.trace.data.spans) == num_threads\n-\n-\n-def test_traces_buffer_expires_after_ttl(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_TRACE_BUFFER_TTL_SECONDS\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- time.sleep(1)\n-\n- assert request_id_1 not in trace_manager._traces\n-\n- # Clear singleton instance again to avoid side effects to other tests\n- InMemoryTraceManager._instance = None\n-\n-\n-def test_traces_buffer_expires_and_log_when_timeout_is_set(monkeypatch):\n- # Setting MLFLOW_TRACE_TIMEOUT_SECONDS let MLflow to periodically check the\n- # expired traces and log expired ones to the backend.\n- monkeypatch.setenv(\"MLFLOW_TRACE_TIMEOUT_SECONDS\", \"1\")\n- monkeypatch.setenv(\"MLFLOW_TRACE_TTL_CHECK_INTERVAL_SECONDS\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_info = create_test_trace_info(request_id_1, \"test\")\n- trace_manager.register_trace(12345, trace_info)\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert trace_manager._traces.get(request_id_1) is not None\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- assert request_id_1 in trace_manager._traces\n-\n- time.sleep(3)\n-\n- assert request_id_1 not in trace_manager._traces\n- assert span_1_1.status.status_code == SpanStatusCode.ERROR\n-\n-\n-def test_traces_buffer_max_size_limit(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_TRACE_BUFFER_MAX_SIZE\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"experiment\"))\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces) == 1\n-\n- request_id_2 = \"tr-2\"\n- trace_manager.register_trace(67890, create_test_trace_info(request_id_2, \"experiment\"))\n- span_2_1 = _create_test_span(request_id_2)\n- trace_manager.register_span(span_2_1)\n-\n- assert request_id_1 not in trace_manager._traces\n- assert request_id_2 in trace_manager._traces\n- assert len(trace_manager._traces) == 1\n-\n- # Clear singleton instance again to avoid side effects to other tests\n- InMemoryTraceManager._instance = None\n-\n-\n-def test_get_span_from_id():\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- span_1_1 = _create_test_span(request_id_1, trace_id=111, span_id=1)\n- span_1_2 = _create_test_span(request_id_1, trace_id=111, span_id=2, parent_id=1)\n-\n- request_id_2 = \"tr-2\"\n- trace_manager.register_trace(67890, create_test_trace_info(request_id_2, \"test\"))\n- span_2_1 = _create_test_span(request_id_2, trace_id=222, span_id=1)\n- span_2_2 = _create_test_span(request_id_2, trace_id=222, span_id=2, parent_id=1)\n-\n- # Add a span for a new trace\n- trace_manager.register_span(span_1_1)\n- trace_manager.register_span(span_1_2)\n- trace_manager.register_span(span_2_1)\n- trace_manager.register_span(span_2_2)\n-\n- assert trace_manager.get_span_from_id(request_id_1, span_1_1.span_id) == span_1_1\n- assert trace_manager.get_span_from_id(request_id_2, span_2_2.span_id) == span_2_2\n-\n-\n-def test_get_root_span_id():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- span_1_1 = _create_test_span(request_id_1, span_id=1)\n- span_1_2 = _create_test_span(request_id_1, span_id=2, parent_id=1)\n-\n- # Add a span for a new trace\n- trace_manager.register_span(span_1_1)\n- trace_manager.register_span(span_1_2)\n-\n- assert trace_manager.get_root_span_id(request_id_1) == span_1_1.span_id\n-\n- # Non-existing trace\n- assert trace_manager.get_root_span_id(\"tr-2\") is None\n-\n-\n-def test_register_prompt():\n- prompt_version = PromptVersion(name=\"test_prompt\", version=1, template=\"Test template\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- trace_manager.register_prompt(request_id_1, prompt_version)\n-\n- trace = trace_manager._traces[request_id_1]\n- assert trace.prompts == [prompt_version]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"}\n- ]\n-\n- # Adding multiple prompts to the same trace\n- prompt_version_2 = PromptVersion(name=\"test_prompt_2\", version=2, template=\"Test template 2\")\n-\n- trace_manager.register_prompt(request_id_1, prompt_version_2)\n- assert trace.prompts == [prompt_version, prompt_version_2]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"},\n- {\"name\": \"test_prompt_2\", \"version\": \"2\"},\n- ]\n-\n- # Registering the same prompt should not add it again\n- trace_manager.register_prompt(request_id_1, prompt_version)\n- assert trace.prompts == [prompt_version, prompt_version_2]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"},\n- {\"name\": \"test_prompt_2\", \"version\": \"2\"},\n- ]\n-\n-\n-def test_register_prompt_thread_safety():\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- prompt_versions = [\n- PromptVersion(name=f\"test_prompt_{i}\", version=i, template=f\"Test template {i}\")\n- for i in range(10)\n- ]\n-\n- def register_prompt(prompt_version):\n- trace_manager.register_prompt(request_id_1, prompt_version)\n-\n- threads = [Thread(target=register_prompt, args=[pv]) for pv in prompt_versions]\n- for thread in threads:\n- thread.start()\n- for thread in threads:\n- thread.join()\n-\n- trace = trace_manager._traces[request_id_1]\n- assert len(trace.prompts) == len(prompt_versions)\n- assert len(json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY])) == len(prompt_versions)\n-\n-\n-def _create_test_span(\n- request_id=\"tr-12345\",\n- trace_id: int = 12345,\n- span_id: int = 123,\n- parent_id: int | None = None,\n- start_time=None,\n- end_time=None,\n-):\n- mock_otel_span = create_mock_otel_span(\n- trace_id=trace_id,\n- span_id=span_id,\n- parent_id=parent_id,\n- start_time=start_time,\n- end_time=end_time,\n- )\n-\n- span = LiveSpan(mock_otel_span, request_id)\n- span.set_status(\"OK\")\n- return span\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S1\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n },\n {\n \"spec_id\": \"S2\",\n \"shared_functions\": 0,\n \"shared_lines\": 0\n }\n ]\n}"}}},{"rowIdx":99,"cells":{"schema_version":{"kind":"number","value":3,"string":"3"},"instance_id":{"kind":"string","value":"mlflow__mlflow.93dab383.test_trace_manager.bb95fbcd.lv1"},"spec":{"kind":"string","value":"{\n \"spec_id\": \"S20\",\n \"text\": \"When updating the tag, if a current tag value exists, it must be parsed as JSON and validated to be a list type. Invalid JSON or non-list types should raise an MlflowException with a descriptive error message.\",\n \"category\": \"behavioral\",\n \"code_mapping\": [\n {\n \"file\": \"mlflow/tracing/utils/prompt.py\",\n \"lines\": \"22-35\",\n \"function\": \"update_linked_prompts_tag\"\n }\n ]\n}"},"mask_patch":{"kind":"string","value":"diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py\nindex 90d90ab25..4b2aceada 100644\n--- a/mlflow/entities/span.py\n+++ b/mlflow/entities/span.py\n@@ -117,10 +117,10 @@ class Span:\n \"\"\"The trace ID of the span, a unique identifier for the trace it belongs to.\"\"\"\n return self.get_attribute(SpanAttributeKey.REQUEST_ID)\n \n- @property\n- def request_id(self) -> str:\n- \"\"\"Deprecated. Use `trace_id` instead.\"\"\"\n- return self.trace_id\n+\n+\n+\n+\n \n @property\n def span_id(self) -> str:\n@@ -212,17 +212,17 @@ class Span:\n f\"span_id={self.span_id!r}, parent_id={self.parent_id!r})\"\n )\n \n- def get_attribute(self, key: str) -> Any | None:\n- \"\"\"\n- Get a single attribute value from the span.\n \n- Args:\n- key: The key of the attribute to get.\n \n- Returns:\n- The value of the attribute if it exists, otherwise None.\n- \"\"\"\n- return self._attributes.get(key)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def to_dict(self) -> dict[str, Any]:\n return {\n@@ -665,14 +665,14 @@ class LiveSpan(Span):\n def from_dict(cls, data: dict[str, Any]) -> \"Span\":\n raise NotImplementedError(\"The `from_dict` method is not supported for the LiveSpan class.\")\n \n- def to_immutable_span(self) -> \"Span\":\n- \"\"\"\n- Downcast the live span object to the immutable span.\n \n- :meta private:\n- \"\"\"\n- # All state of the live span is already persisted in the OpenTelemetry span object.\n- return Span(self._span)\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def from_immutable_span(\n@@ -781,16 +781,16 @@ class NoOpSpan(Span):\n self._span = NonRecordingSpan(context=None)\n self._attributes = {}\n \n- @property\n- def trace_id(self):\n- \"\"\"\n- No-op span returns a special trace ID to distinguish it from the real spans.\n- \"\"\"\n- return NO_OP_SPAN_TRACE_ID\n \n- @property\n- def span_id(self):\n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @property\n def name(self):\n@@ -808,9 +808,9 @@ class NoOpSpan(Span):\n def context(self):\n return None\n \n- @property\n- def parent_id(self):\n- return None\n+\n+\n+\n \n @property\n def status(self):\n@@ -898,11 +898,11 @@ class _CachedSpanAttributesRegistry(_SpanAttributesRegistry):\n spans that are immutable, and thus implemented as a subclass of _SpanAttributesRegistry.\n \"\"\"\n \n- @lru_cache(maxsize=128)\n- def get(self, key: str):\n- return super().get(key)\n+\n+\n+\n \n def set(self, key: str, value: Any):\n raise MlflowException(\n \"The attributes of the immutable span must not be updated.\", INVALID_PARAMETER_VALUE\n- )\n+ )\n\\ No newline at end of file\ndiff --git a/mlflow/entities/trace_data.py b/mlflow/entities/trace_data.py\nindex ed3d862b4..69ed4e456 100644\n--- a/mlflow/entities/trace_data.py\n+++ b/mlflow/entities/trace_data.py\n@@ -19,8 +19,8 @@ class TraceData:\n \n # NB: Custom constructor to allow passing additional kwargs for backward compatibility for\n # DBX agent evaluator. Once they migrates to trace V3 schema, we can remove this.\n- def __init__(self, spans: list[Span] | None = None, **kwargs):\n- self.spans = spans or []\n+\n+\n \n @classmethod\n def from_dict(cls, d):\n@@ -64,22 +64,19 @@ class TraceData:\n result[span_name] = span.outputs\n return result\n \n- def _get_root_span(self) -> Span | None:\n- for span in self.spans:\n- if span.parent_id is None:\n- return span\n+\n+\n+\n+\n \n # `request` and `response` are preserved for backward compatibility with v2\n- @property\n- def request(self) -> str | None:\n- if span := self._get_root_span():\n- # Accessing the OTel span directly get serialized value directly.\n- return span._span.attributes.get(SpanAttributeKey.INPUTS)\n- return None\n \n- @property\n- def response(self) -> str | None:\n- if span := self._get_root_span():\n- # Accessing the OTel span directly get serialized value directly.\n- return span._span.attributes.get(SpanAttributeKey.OUTPUTS)\n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1b0edf169 100644\n--- a/mlflow/store/model_registry/abstract_store.py\n+++ b/mlflow/store/model_registry/abstract_store.py\n@@ -32,7 +32,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.store.entities.paged_list import PagedList\n-from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n+\n from mlflow.utils.annotations import developer_stable\n from mlflow.utils.logging_utils import eprint\n \ndiff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py\nindex 9c9caa149..94ee05f6e 100644\n--- a/mlflow/tracing/trace_manager.py\n+++ b/mlflow/tracing/trace_manager.py\n@@ -8,9 +8,9 @@ from mlflow.entities import LiveSpan, Trace, TraceData, TraceInfo\n from mlflow.entities.model_registry import PromptVersion\n from mlflow.environment_variables import MLFLOW_TRACE_TIMEOUT_SECONDS\n from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n-from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n-from mlflow.tracing.utils.timeout import get_trace_cache_with_timeout\n-from mlflow.tracing.utils.truncation import set_request_response_preview\n+\n+\n+\n \n _logger = logging.getLogger(__name__)\n \n@@ -23,14 +23,14 @@ class _Trace:\n span_dict: dict[str, LiveSpan] = field(default_factory=dict)\n prompts: list[PromptVersion] = field(default_factory=list)\n \n- def to_mlflow_trace(self) -> Trace:\n- trace_data = TraceData()\n- for span in self.span_dict.values():\n- # Convert LiveSpan, mutable objects, into immutable Span objects before persisting.\n- trace_data.spans.append(span.to_immutable_span())\n \n- set_request_response_preview(self.info, trace_data)\n- return Trace(self.info, trace_data)\n+\n+\n+\n+\n+\n+\n+\n \n def get_root_span(self) -> LiveSpan | None:\n for span in self.span_dict.values():\n@@ -57,73 +57,73 @@ class InMemoryTraceManager:\n _instance_lock = threading.RLock()\n _instance = None\n \n- @classmethod\n- def get_instance(cls):\n- if cls._instance is None:\n- with cls._instance_lock:\n- if cls._instance is None:\n- cls._instance = InMemoryTraceManager()\n- return cls._instance\n-\n- def __init__(self):\n- # In-memory cache to store trace_od -> _Trace mapping.\n- self._traces = get_trace_cache_with_timeout()\n-\n- # Store mapping between OpenTelemetry trace ID and MLflow trace ID\n- self._otel_id_to_mlflow_trace_id: dict[int, str] = {}\n- self._lock = threading.RLock() # Lock for _traces\n-\n- def register_trace(self, otel_trace_id: int, trace_info: TraceInfo):\n- \"\"\"\n- Register a new trace info object to the in-memory trace registry.\n \n- Args:\n- otel_trace_id: The OpenTelemetry trace ID for the new trace.\n- trace_info: The trace info object to be stored.\n- \"\"\"\n- # Check for a new timeout setting whenever a new trace is created.\n- self._check_timeout_update()\n- with self._lock:\n- self._traces[trace_info.trace_id] = _Trace(trace_info)\n- self._otel_id_to_mlflow_trace_id[otel_trace_id] = trace_info.trace_id\n \n- def register_span(self, span: LiveSpan):\n- \"\"\"\n- Store the given span in the in-memory trace data.\n \n- Args:\n- span: The span to be stored.\n- \"\"\"\n- if not isinstance(span, LiveSpan):\n- _logger.debug(f\"Invalid span object {type(span)} is passed. Skipping.\")\n- return\n \n- with self._lock:\n- trace_data_dict = self._traces[span.request_id].span_dict\n- trace_data_dict[span.span_id] = span\n \n- def register_prompt(self, trace_id: str, prompt: PromptVersion):\n- \"\"\"\n- Register a prompt to link to the trace with the given trace ID.\n \n- Args:\n- trace_id: The ID of the trace to which the prompt belongs.\n- prompt: The prompt version to be registered.\n- \"\"\"\n- with self._lock:\n- if prompt not in self._traces[trace_id].prompts:\n- self._traces[trace_id].prompts.append(prompt)\n-\n- # NB: Set prompt URIs in trace tags for linking. This is a short-term solution until\n- # LinkPromptsToTraces endpoint is implemented in the backend.\n- # TODO: Remove this once LinkPromptsToTraces endpoint is implemented in the backend.\n- try:\n- current_tag = self._traces[trace_id].info.tags.get(LINKED_PROMPTS_TAG_KEY)\n- updated_tag = update_linked_prompts_tag(current_tag, [prompt])\n- self._traces[trace_id].info.tags[LINKED_PROMPTS_TAG_KEY] = updated_tag\n- except Exception:\n- _logger.debug(f\"Failed to update prompts tag for trace {trace_id}\", exc_info=True)\n- raise\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @contextlib.contextmanager\n def get_trace(self, trace_id: str) -> Generator[_Trace | None, None, None]:\n@@ -135,28 +135,28 @@ class InMemoryTraceManager:\n with self._lock:\n yield self._traces.get(trace_id)\n \n- def get_span_from_id(self, trace_id: str, span_id: str) -> LiveSpan | None:\n- \"\"\"\n- Get a span object for the given trace_id and span_id.\n- \"\"\"\n- with self._lock:\n- trace = self._traces.get(trace_id)\n \n- return trace.span_dict.get(span_id) if trace else None\n \n- def get_root_span_id(self, trace_id) -> str | None:\n- \"\"\"\n- Get the root span ID for the given trace ID.\n- \"\"\"\n- with self._lock:\n- trace = self._traces.get(trace_id)\n \n- if trace:\n- for span in trace.span_dict.values():\n- if span.parent_id is None:\n- return span.span_id\n \n- return None\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n def get_mlflow_trace_id_from_otel_id(self, otel_trace_id: int) -> str | None:\n \"\"\"\n@@ -172,38 +172,38 @@ class InMemoryTraceManager:\n if trace:\n trace.info.trace_metadata[key] = value\n \n- def pop_trace(self, otel_trace_id: int) -> ManagerTrace | None:\n- \"\"\"\n- Pop trace data for the given OpenTelemetry trace ID and\n- return it as a ManagerTrace wrapper containing the trace and prompts.\n- \"\"\"\n- with self._lock:\n- mlflow_trace_id = self._otel_id_to_mlflow_trace_id.pop(otel_trace_id, None)\n- internal_trace = self._traces.pop(mlflow_trace_id, None) if mlflow_trace_id else None\n- if internal_trace is None:\n- return None\n- return ManagerTrace(\n- trace=internal_trace.to_mlflow_trace(), prompts=internal_trace.prompts\n- )\n-\n- def _check_timeout_update(self):\n- \"\"\"\n- TTL/Timeout may be updated by users after initial cache creation. This method checks\n- for the update and create a new cache instance with the updated timeout.\n- \"\"\"\n- new_timeout = MLFLOW_TRACE_TIMEOUT_SECONDS.get()\n- if new_timeout != getattr(self._traces, \"timeout\", None):\n- if len(self._traces) > 0:\n- _logger.warning(\n- f\"The timeout of the trace buffer has been updated to {new_timeout} seconds. \"\n- \"This operation discards all in-progress traces at the moment. Please make \"\n- \"sure to update the timeout when there are no in-progress traces.\"\n- )\n-\n- with self._lock:\n- # We need to check here again in case this method runs in parallel\n- if new_timeout != getattr(self._traces, \"timeout\", None):\n- self._traces = get_trace_cache_with_timeout()\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n @classmethod\n def reset(cls):\n@@ -211,4 +211,4 @@ class InMemoryTraceManager:\n if cls._instance:\n with cls._instance._lock:\n cls._instance._traces.clear()\n- cls._instance = None\n+ cls._instance = None\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/prompt.py b/mlflow/tracing/utils/prompt.py\nindex 8a9847515..1200f9638 100644\n--- a/mlflow/tracing/utils/prompt.py\n+++ b/mlflow/tracing/utils/prompt.py\n@@ -6,42 +6,40 @@ from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n \n \n # TODO: Remove tag based linking once we migrate to LinkPromptsToTraces endpoint\n-def update_linked_prompts_tag(current_tag_value: str | None, prompt_versions: list[PromptVersion]):\n- \"\"\"\n- Utility method to update linked prompts tag value with a new prompt version.\n-\n- Args:\n- current_tag_value: Current JSON string value of the linked prompts tag\n- prompt_versions: List of PromptVersion objects to add\n-\n- Returns:\n- Updated JSON string with new entries added (avoiding duplicates)\n-\n- Raises:\n- MlflowException: If current tag value has invalid JSON or format\n- \"\"\"\n- if current_tag_value is not None:\n- try:\n- parsed_prompts_tag_value = json.loads(current_tag_value)\n- if not isinstance(parsed_prompts_tag_value, list):\n- raise MlflowException(\n- f\"Invalid format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n- )\n- except json.JSONDecodeError:\n- raise MlflowException(\n- f\"Invalid JSON format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n- )\n- else:\n- parsed_prompts_tag_value = []\n-\n- new_prompt_entries = [\n- {\"name\": prompt_version.name, \"version\": str(prompt_version.version)}\n- for prompt_version in prompt_versions\n- ]\n-\n- prompts_to_add = [p for p in new_prompt_entries if p not in parsed_prompts_tag_value]\n- if not prompts_to_add:\n- return current_tag_value\n-\n- parsed_prompts_tag_value.extend(prompts_to_add)\n- return json.dumps(parsed_prompts_tag_value)\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\ndiff --git a/mlflow/tracing/utils/timeout.py b/mlflow/tracing/utils/timeout.py\nindex dcd7c8b71..8eb5847c5 100644\n--- a/mlflow/tracing/utils/timeout.py\n+++ b/mlflow/tracing/utils/timeout.py\n@@ -25,28 +25,28 @@ _TRACE_EXPIRATION_MSG = (\n )\n \n \n-def get_trace_cache_with_timeout() -> Cache:\n- \"\"\"\n- Return a cache object that stores traces in-memory while they are in-progress.\n \n- If the timeout is specified, this returns a customized cache that logs the\n- expired traces to the backend. Otherwise, this returns a regular cache.\n- \"\"\"\n \n- if timeout := MLFLOW_TRACE_TIMEOUT_SECONDS.get():\n- return MlflowTraceTimeoutCache(\n- timeout=timeout,\n- maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n- )\n \n- # NB: Ideally we should return the vanilla Cache object only with maxsize.\n- # But we used TTLCache before introducing the timeout feature (that does not\n- # monitor timeout periodically nor log the expired traces). To keep the\n- # backward compatibility, we return TTLCache.\n- return TTLCache(\n- ttl=MLFLOW_TRACE_BUFFER_TTL_SECONDS.get(),\n- maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n- )\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n class _TimedCache(Cache):\n@@ -247,4 +247,4 @@ class MlflowTraceTimeoutCache(_TimedCache):\n def clear(self):\n super().clear()\n self._expire_checker_stop_event.set()\n- self._expire_checker_thread.join()\n+ self._expire_checker_thread.join()\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/truncation.py b/mlflow/tracing/utils/truncation.py\nindex 2d69c9043..3b6b4c0c9 100644\n--- a/mlflow/tracing/utils/truncation.py\n+++ b/mlflow/tracing/utils/truncation.py\n@@ -12,49 +12,49 @@ from mlflow.tracking._tracking_service.utils import get_tracking_uri\n from mlflow.utils.uri import is_databricks_uri\n \n \n-def set_request_response_preview(trace_info: TraceInfo, trace_data: TraceData) -> None:\n- \"\"\"\n- Set the request and response previews for the trace info.\n- \"\"\"\n- # If request/response preview is already set by users via `mlflow.update_current_trace`,\n- # we don't override it with the truncated version.\n- if trace_info.request_preview is None:\n- trace_info.request_preview = _get_truncated_preview(trace_data.request, role=\"user\")\n- if trace_info.response_preview is None:\n- trace_info.response_preview = _get_truncated_preview(trace_data.response, role=\"assistant\")\n \n \n-def _get_truncated_preview(request_or_response: str | dict[str, Any] | None, role: str) -> str:\n- \"\"\"\n- Truncate the request preview to fit the max length.\n- \"\"\"\n- if request_or_response is None:\n- return \"\"\n-\n- max_length = _get_max_length()\n-\n- content = None\n- obj = None\n- if isinstance(request_or_response, dict):\n- obj = request_or_response\n- request_or_response = json.dumps(request_or_response)\n- elif isinstance(request_or_response, str):\n- try:\n- obj = json.loads(request_or_response)\n- except json.JSONDecodeError:\n- pass\n-\n- if obj is not None:\n- if messages := _try_extract_messages(obj):\n- msg = _get_last_message(messages, role=role)\n- content = _get_text_content_from_message(msg)\n-\n- content = content or request_or_response\n-\n- if len(content) <= max_length:\n- return content\n \n- return content[: max_length - 3] + \"...\"\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n+\n \n \n @lru_cache(maxsize=1)\n@@ -127,4 +127,4 @@ def _get_text_content_from_message(message: dict[str, Any]) -> str:\n return part.get(\"text\")\n elif isinstance(content, str):\n return content\n- return \"\"\n+ return \"\"\n\\ No newline at end of file\n"},"gold_patch":{"kind":"string","value":"diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py\nindex 90d90ab25..4b2aceada 100644\n+++ a/mlflow/entities/span.py\n--- b/mlflow/entities/span.py\n@@ -117,10 +117,10 @@ class Span:\n \"\"\"The trace ID of the span, a unique identifier for the trace it belongs to.\"\"\"\n return self.get_attribute(SpanAttributeKey.REQUEST_ID)\n \n+ @property\n+ def request_id(self) -> str:\n+ \"\"\"Deprecated. Use `trace_id` instead.\"\"\"\n+ return self.trace_id\n-\n-\n-\n-\n \n @property\n def span_id(self) -> str:\n@@ -212,17 +212,17 @@ class Span:\n f\"span_id={self.span_id!r}, parent_id={self.parent_id!r})\"\n )\n \n+ def get_attribute(self, key: str) -> Any | None:\n+ \"\"\"\n+ Get a single attribute value from the span.\n \n+ Args:\n+ key: The key of the attribute to get.\n \n+ Returns:\n+ The value of the attribute if it exists, otherwise None.\n+ \"\"\"\n+ return self._attributes.get(key)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def to_dict(self) -> dict[str, Any]:\n return {\n@@ -665,14 +665,14 @@ class LiveSpan(Span):\n def from_dict(cls, data: dict[str, Any]) -> \"Span\":\n raise NotImplementedError(\"The `from_dict` method is not supported for the LiveSpan class.\")\n \n+ def to_immutable_span(self) -> \"Span\":\n+ \"\"\"\n+ Downcast the live span object to the immutable span.\n \n+ :meta private:\n+ \"\"\"\n+ # All state of the live span is already persisted in the OpenTelemetry span object.\n+ return Span(self._span)\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def from_immutable_span(\n@@ -781,16 +781,16 @@ class NoOpSpan(Span):\n self._span = NonRecordingSpan(context=None)\n self._attributes = {}\n \n+ @property\n+ def trace_id(self):\n+ \"\"\"\n+ No-op span returns a special trace ID to distinguish it from the real spans.\n+ \"\"\"\n+ return NO_OP_SPAN_TRACE_ID\n \n+ @property\n+ def span_id(self):\n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @property\n def name(self):\n@@ -808,9 +808,9 @@ class NoOpSpan(Span):\n def context(self):\n return None\n \n+ @property\n+ def parent_id(self):\n+ return None\n-\n-\n-\n \n @property\n def status(self):\n@@ -898,11 +898,11 @@ class _CachedSpanAttributesRegistry(_SpanAttributesRegistry):\n spans that are immutable, and thus implemented as a subclass of _SpanAttributesRegistry.\n \"\"\"\n \n+ @lru_cache(maxsize=128)\n+ def get(self, key: str):\n+ return super().get(key)\n-\n-\n-\n \n def set(self, key: str, value: Any):\n raise MlflowException(\n \"The attributes of the immutable span must not be updated.\", INVALID_PARAMETER_VALUE\n+ )\n- )\n\\ No newline at end of file\ndiff --git a/mlflow/entities/trace_data.py b/mlflow/entities/trace_data.py\nindex ed3d862b4..69ed4e456 100644\n+++ a/mlflow/entities/trace_data.py\n--- b/mlflow/entities/trace_data.py\n@@ -19,8 +19,8 @@ class TraceData:\n \n # NB: Custom constructor to allow passing additional kwargs for backward compatibility for\n # DBX agent evaluator. Once they migrates to trace V3 schema, we can remove this.\n+ def __init__(self, spans: list[Span] | None = None, **kwargs):\n+ self.spans = spans or []\n-\n-\n \n @classmethod\n def from_dict(cls, d):\n@@ -64,19 +64,22 @@ class TraceData:\n result[span_name] = span.outputs\n return result\n \n+ def _get_root_span(self) -> Span | None:\n+ for span in self.spans:\n+ if span.parent_id is None:\n+ return span\n-\n-\n-\n-\n \n # `request` and `response` are preserved for backward compatibility with v2\n+ @property\n+ def request(self) -> str | None:\n+ if span := self._get_root_span():\n+ # Accessing the OTel span directly get serialized value directly.\n+ return span._span.attributes.get(SpanAttributeKey.INPUTS)\n+ return None\n \n+ @property\n+ def response(self) -> str | None:\n+ if span := self._get_root_span():\n+ # Accessing the OTel span directly get serialized value directly.\n+ return span._span.attributes.get(SpanAttributeKey.OUTPUTS)\n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/store/model_registry/abstract_store.py b/mlflow/store/model_registry/abstract_store.py\nindex 9da9d29a4..1b0edf169 100644\n+++ a/mlflow/store/model_registry/abstract_store.py\n--- b/mlflow/store/model_registry/abstract_store.py\n@@ -32,7 +32,7 @@ from mlflow.protos.databricks_pb2 import (\n ErrorCode,\n )\n from mlflow.store.entities.paged_list import PagedList\n+from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n-\n from mlflow.utils.annotations import developer_stable\n from mlflow.utils.logging_utils import eprint\n \ndiff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py\nindex 9c9caa149..94ee05f6e 100644\n+++ a/mlflow/tracing/trace_manager.py\n--- b/mlflow/tracing/trace_manager.py\n@@ -8,9 +8,9 @@ from mlflow.entities import LiveSpan, Trace, TraceData, TraceInfo\n from mlflow.entities.model_registry import PromptVersion\n from mlflow.environment_variables import MLFLOW_TRACE_TIMEOUT_SECONDS\n from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n+from mlflow.tracing.utils.prompt import update_linked_prompts_tag\n+from mlflow.tracing.utils.timeout import get_trace_cache_with_timeout\n+from mlflow.tracing.utils.truncation import set_request_response_preview\n-\n-\n-\n \n _logger = logging.getLogger(__name__)\n \n@@ -23,14 +23,14 @@ class _Trace:\n span_dict: dict[str, LiveSpan] = field(default_factory=dict)\n prompts: list[PromptVersion] = field(default_factory=list)\n \n+ def to_mlflow_trace(self) -> Trace:\n+ trace_data = TraceData()\n+ for span in self.span_dict.values():\n+ # Convert LiveSpan, mutable objects, into immutable Span objects before persisting.\n+ trace_data.spans.append(span.to_immutable_span())\n \n+ set_request_response_preview(self.info, trace_data)\n+ return Trace(self.info, trace_data)\n-\n-\n-\n-\n-\n-\n-\n \n def get_root_span(self) -> LiveSpan | None:\n for span in self.span_dict.values():\n@@ -57,73 +57,73 @@ class InMemoryTraceManager:\n _instance_lock = threading.RLock()\n _instance = None\n \n+ @classmethod\n+ def get_instance(cls):\n+ if cls._instance is None:\n+ with cls._instance_lock:\n+ if cls._instance is None:\n+ cls._instance = InMemoryTraceManager()\n+ return cls._instance\n+\n+ def __init__(self):\n+ # In-memory cache to store trace_od -> _Trace mapping.\n+ self._traces = get_trace_cache_with_timeout()\n+\n+ # Store mapping between OpenTelemetry trace ID and MLflow trace ID\n+ self._otel_id_to_mlflow_trace_id: dict[int, str] = {}\n+ self._lock = threading.RLock() # Lock for _traces\n+\n+ def register_trace(self, otel_trace_id: int, trace_info: TraceInfo):\n+ \"\"\"\n+ Register a new trace info object to the in-memory trace registry.\n \n+ Args:\n+ otel_trace_id: The OpenTelemetry trace ID for the new trace.\n+ trace_info: The trace info object to be stored.\n+ \"\"\"\n+ # Check for a new timeout setting whenever a new trace is created.\n+ self._check_timeout_update()\n+ with self._lock:\n+ self._traces[trace_info.trace_id] = _Trace(trace_info)\n+ self._otel_id_to_mlflow_trace_id[otel_trace_id] = trace_info.trace_id\n \n+ def register_span(self, span: LiveSpan):\n+ \"\"\"\n+ Store the given span in the in-memory trace data.\n \n+ Args:\n+ span: The span to be stored.\n+ \"\"\"\n+ if not isinstance(span, LiveSpan):\n+ _logger.debug(f\"Invalid span object {type(span)} is passed. Skipping.\")\n+ return\n \n+ with self._lock:\n+ trace_data_dict = self._traces[span.request_id].span_dict\n+ trace_data_dict[span.span_id] = span\n \n+ def register_prompt(self, trace_id: str, prompt: PromptVersion):\n+ \"\"\"\n+ Register a prompt to link to the trace with the given trace ID.\n \n+ Args:\n+ trace_id: The ID of the trace to which the prompt belongs.\n+ prompt: The prompt version to be registered.\n+ \"\"\"\n+ with self._lock:\n+ if prompt not in self._traces[trace_id].prompts:\n+ self._traces[trace_id].prompts.append(prompt)\n+\n+ # NB: Set prompt URIs in trace tags for linking. This is a short-term solution until\n+ # LinkPromptsToTraces endpoint is implemented in the backend.\n+ # TODO: Remove this once LinkPromptsToTraces endpoint is implemented in the backend.\n+ try:\n+ current_tag = self._traces[trace_id].info.tags.get(LINKED_PROMPTS_TAG_KEY)\n+ updated_tag = update_linked_prompts_tag(current_tag, [prompt])\n+ self._traces[trace_id].info.tags[LINKED_PROMPTS_TAG_KEY] = updated_tag\n+ except Exception:\n+ _logger.debug(f\"Failed to update prompts tag for trace {trace_id}\", exc_info=True)\n+ raise\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @contextlib.contextmanager\n def get_trace(self, trace_id: str) -> Generator[_Trace | None, None, None]:\n@@ -135,28 +135,28 @@ class InMemoryTraceManager:\n with self._lock:\n yield self._traces.get(trace_id)\n \n+ def get_span_from_id(self, trace_id: str, span_id: str) -> LiveSpan | None:\n+ \"\"\"\n+ Get a span object for the given trace_id and span_id.\n+ \"\"\"\n+ with self._lock:\n+ trace = self._traces.get(trace_id)\n \n+ return trace.span_dict.get(span_id) if trace else None\n \n+ def get_root_span_id(self, trace_id) -> str | None:\n+ \"\"\"\n+ Get the root span ID for the given trace ID.\n+ \"\"\"\n+ with self._lock:\n+ trace = self._traces.get(trace_id)\n \n+ if trace:\n+ for span in trace.span_dict.values():\n+ if span.parent_id is None:\n+ return span.span_id\n \n+ return None\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n def get_mlflow_trace_id_from_otel_id(self, otel_trace_id: int) -> str | None:\n \"\"\"\n@@ -172,38 +172,38 @@ class InMemoryTraceManager:\n if trace:\n trace.info.trace_metadata[key] = value\n \n+ def pop_trace(self, otel_trace_id: int) -> ManagerTrace | None:\n+ \"\"\"\n+ Pop trace data for the given OpenTelemetry trace ID and\n+ return it as a ManagerTrace wrapper containing the trace and prompts.\n+ \"\"\"\n+ with self._lock:\n+ mlflow_trace_id = self._otel_id_to_mlflow_trace_id.pop(otel_trace_id, None)\n+ internal_trace = self._traces.pop(mlflow_trace_id, None) if mlflow_trace_id else None\n+ if internal_trace is None:\n+ return None\n+ return ManagerTrace(\n+ trace=internal_trace.to_mlflow_trace(), prompts=internal_trace.prompts\n+ )\n+\n+ def _check_timeout_update(self):\n+ \"\"\"\n+ TTL/Timeout may be updated by users after initial cache creation. This method checks\n+ for the update and create a new cache instance with the updated timeout.\n+ \"\"\"\n+ new_timeout = MLFLOW_TRACE_TIMEOUT_SECONDS.get()\n+ if new_timeout != getattr(self._traces, \"timeout\", None):\n+ if len(self._traces) > 0:\n+ _logger.warning(\n+ f\"The timeout of the trace buffer has been updated to {new_timeout} seconds. \"\n+ \"This operation discards all in-progress traces at the moment. Please make \"\n+ \"sure to update the timeout when there are no in-progress traces.\"\n+ )\n+\n+ with self._lock:\n+ # We need to check here again in case this method runs in parallel\n+ if new_timeout != getattr(self._traces, \"timeout\", None):\n+ self._traces = get_trace_cache_with_timeout()\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n @classmethod\n def reset(cls):\n@@ -211,4 +211,4 @@ class InMemoryTraceManager:\n if cls._instance:\n with cls._instance._lock:\n cls._instance._traces.clear()\n+ cls._instance = None\n- cls._instance = None\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/prompt.py b/mlflow/tracing/utils/prompt.py\nindex 8a9847515..1200f9638 100644\n+++ a/mlflow/tracing/utils/prompt.py\n--- b/mlflow/tracing/utils/prompt.py\n@@ -6,40 +6,42 @@ from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n \n \n # TODO: Remove tag based linking once we migrate to LinkPromptsToTraces endpoint\n+def update_linked_prompts_tag(current_tag_value: str | None, prompt_versions: list[PromptVersion]):\n+ \"\"\"\n+ Utility method to update linked prompts tag value with a new prompt version.\n+\n+ Args:\n+ current_tag_value: Current JSON string value of the linked prompts tag\n+ prompt_versions: List of PromptVersion objects to add\n+\n+ Returns:\n+ Updated JSON string with new entries added (avoiding duplicates)\n+\n+ Raises:\n+ MlflowException: If current tag value has invalid JSON or format\n+ \"\"\"\n+ if current_tag_value is not None:\n+ try:\n+ parsed_prompts_tag_value = json.loads(current_tag_value)\n+ if not isinstance(parsed_prompts_tag_value, list):\n+ raise MlflowException(\n+ f\"Invalid format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n+ )\n+ except json.JSONDecodeError:\n+ raise MlflowException(\n+ f\"Invalid JSON format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n+ )\n+ else:\n+ parsed_prompts_tag_value = []\n+\n+ new_prompt_entries = [\n+ {\"name\": prompt_version.name, \"version\": str(prompt_version.version)}\n+ for prompt_version in prompt_versions\n+ ]\n+\n+ prompts_to_add = [p for p in new_prompt_entries if p not in parsed_prompts_tag_value]\n+ if not prompts_to_add:\n+ return current_tag_value\n+\n+ parsed_prompts_tag_value.extend(prompts_to_add)\n+ return json.dumps(parsed_prompts_tag_value)\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\ndiff --git a/mlflow/tracing/utils/timeout.py b/mlflow/tracing/utils/timeout.py\nindex dcd7c8b71..8eb5847c5 100644\n+++ a/mlflow/tracing/utils/timeout.py\n--- b/mlflow/tracing/utils/timeout.py\n@@ -25,28 +25,28 @@ _TRACE_EXPIRATION_MSG = (\n )\n \n \n+def get_trace_cache_with_timeout() -> Cache:\n+ \"\"\"\n+ Return a cache object that stores traces in-memory while they are in-progress.\n \n+ If the timeout is specified, this returns a customized cache that logs the\n+ expired traces to the backend. Otherwise, this returns a regular cache.\n+ \"\"\"\n \n+ if timeout := MLFLOW_TRACE_TIMEOUT_SECONDS.get():\n+ return MlflowTraceTimeoutCache(\n+ timeout=timeout,\n+ maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n+ )\n \n+ # NB: Ideally we should return the vanilla Cache object only with maxsize.\n+ # But we used TTLCache before introducing the timeout feature (that does not\n+ # monitor timeout periodically nor log the expired traces). To keep the\n+ # backward compatibility, we return TTLCache.\n+ return TTLCache(\n+ ttl=MLFLOW_TRACE_BUFFER_TTL_SECONDS.get(),\n+ maxsize=MLFLOW_TRACE_BUFFER_MAX_SIZE.get(),\n+ )\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n class _TimedCache(Cache):\n@@ -247,4 +247,4 @@ class MlflowTraceTimeoutCache(_TimedCache):\n def clear(self):\n super().clear()\n self._expire_checker_stop_event.set()\n+ self._expire_checker_thread.join()\n- self._expire_checker_thread.join()\n\\ No newline at end of file\ndiff --git a/mlflow/tracing/utils/truncation.py b/mlflow/tracing/utils/truncation.py\nindex 2d69c9043..3b6b4c0c9 100644\n+++ a/mlflow/tracing/utils/truncation.py\n--- b/mlflow/tracing/utils/truncation.py\n@@ -12,49 +12,49 @@ from mlflow.tracking._tracking_service.utils import get_tracking_uri\n from mlflow.utils.uri import is_databricks_uri\n \n \n+def set_request_response_preview(trace_info: TraceInfo, trace_data: TraceData) -> None:\n+ \"\"\"\n+ Set the request and response previews for the trace info.\n+ \"\"\"\n+ # If request/response preview is already set by users via `mlflow.update_current_trace`,\n+ # we don't override it with the truncated version.\n+ if trace_info.request_preview is None:\n+ trace_info.request_preview = _get_truncated_preview(trace_data.request, role=\"user\")\n+ if trace_info.response_preview is None:\n+ trace_info.response_preview = _get_truncated_preview(trace_data.response, role=\"assistant\")\n \n \n+def _get_truncated_preview(request_or_response: str | dict[str, Any] | None, role: str) -> str:\n+ \"\"\"\n+ Truncate the request preview to fit the max length.\n+ \"\"\"\n+ if request_or_response is None:\n+ return \"\"\n+\n+ max_length = _get_max_length()\n+\n+ content = None\n+ obj = None\n+ if isinstance(request_or_response, dict):\n+ obj = request_or_response\n+ request_or_response = json.dumps(request_or_response)\n+ elif isinstance(request_or_response, str):\n+ try:\n+ obj = json.loads(request_or_response)\n+ except json.JSONDecodeError:\n+ pass\n+\n+ if obj is not None:\n+ if messages := _try_extract_messages(obj):\n+ msg = _get_last_message(messages, role=role)\n+ content = _get_text_content_from_message(msg)\n+\n+ content = content or request_or_response\n+\n+ if len(content) <= max_length:\n+ return content\n \n+ return content[: max_length - 3] + \"...\"\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n-\n \n \n @lru_cache(maxsize=1)\n@@ -127,4 +127,4 @@ def _get_text_content_from_message(message: dict[str, Any]) -> str:\n return part.get(\"text\")\n elif isinstance(content, str):\n return content\n+ return \"\"\n- return \"\"\n\\ No newline at end of file\n"},"mutant_patch":{"kind":"string","value":"diff --git a/mlflow/tracing/utils/prompt.py b/mlflow/tracing/utils/prompt.py\nindex 8a9847515..3b33fddbd 100644\n--- a/mlflow/tracing/utils/prompt.py\n+++ b/mlflow/tracing/utils/prompt.py\n@@ -23,10 +23,6 @@ def update_linked_prompts_tag(current_tag_value: str | None, prompt_versions: li\n if current_tag_value is not None:\n try:\n parsed_prompts_tag_value = json.loads(current_tag_value)\n- if not isinstance(parsed_prompts_tag_value, list):\n- raise MlflowException(\n- f\"Invalid format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n- )\n except json.JSONDecodeError:\n raise MlflowException(\n f\"Invalid JSON format for '{LINKED_PROMPTS_TAG_KEY}' tag: {current_tag_value}\"\n"},"test_patch":{"kind":"string","value":"diff --git a/tests/tracing/test_trace_manager.py b/tests/tracing/test_trace_manager.py\ndeleted file mode 100644\nindex b5dfcd276..000000000\n--- a/tests/tracing/test_trace_manager.py\n+++ /dev/null\n@@ -1,294 +0,0 @@\n-import json\n-import time\n-from threading import Thread\n-\n-from mlflow.entities import LiveSpan, Span\n-from mlflow.entities.model_registry.prompt_version import PromptVersion\n-from mlflow.entities.span_status import SpanStatusCode\n-from mlflow.prompt.constants import LINKED_PROMPTS_TAG_KEY\n-from mlflow.tracing.trace_manager import InMemoryTraceManager, ManagerTrace\n-\n-from tests.tracing.helper import create_mock_otel_span, create_test_trace_info\n-\n-\n-def test_aggregator_singleton():\n- obj1 = InMemoryTraceManager.get_instance()\n- obj2 = InMemoryTraceManager.get_instance()\n- assert obj1 is obj2\n-\n-\n-def test_add_spans():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- # Add a new trace info\n- request_id_1 = \"tr-1\"\n- trace_id_1 = 12345\n- trace_manager.register_trace(trace_id_1, create_test_trace_info(request_id_1, \"test_1\"))\n-\n- # Add a span for a new trace\n- span_1_1 = _create_test_span(request_id_1, trace_id_1, span_id=1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- # Add more spans to the same trace\n- span_1_1_1 = _create_test_span(request_id_1, trace_id_1, span_id=2, parent_id=1)\n- span_1_1_2 = _create_test_span(request_id_1, trace_id_1, span_id=3, parent_id=1)\n- trace_manager.register_span(span_1_1_1)\n- trace_manager.register_span(span_1_1_2)\n-\n- assert len(trace_manager._traces[request_id_1].span_dict) == 3\n-\n- # Add a span for another trace under the different experiment\n- request_id_2 = \"tr-2\"\n- trace_id_2 = 67890\n- trace_manager.register_trace(trace_id_2, create_test_trace_info(request_id_2, \"test_2\"))\n-\n- span_2_1 = _create_test_span(request_id_2, trace_id_2, span_id=1)\n- span_2_1_1 = _create_test_span(request_id_2, trace_id_2, span_id=2, parent_id=1)\n- trace_manager.register_span(span_2_1)\n- trace_manager.register_span(span_2_1_1)\n-\n- assert request_id_2 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_2].span_dict) == 2\n-\n- # Pop the trace data\n- manager_trace = trace_manager.pop_trace(trace_id_1)\n- assert isinstance(manager_trace, ManagerTrace)\n- assert manager_trace.trace.info.request_id == request_id_1\n- assert len(manager_trace.trace.data.spans) == 3\n- assert request_id_1 not in trace_manager._traces\n-\n- # Convert to MLflow trace to check immutable spans\n- trace = manager_trace.trace\n- assert isinstance(trace.data.spans[0], Span)\n- assert not isinstance(trace.data.spans[0], LiveSpan)\n-\n- manager_trace = trace_manager.pop_trace(trace_id_2)\n- assert isinstance(manager_trace, ManagerTrace)\n- assert manager_trace.trace.info.request_id == request_id_2\n- assert len(manager_trace.trace.data.spans) == 2\n- assert request_id_2 not in trace_manager._traces\n-\n- # Pop a trace that does not exist\n- assert trace_manager.pop_trace(90123) is None\n-\n-\n-def test_add_and_pop_span_thread_safety():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- # Add spans from 10 different threads to 5 different traces\n- trace_ids = list(range(5))\n- num_threads = 10\n-\n- for trace_id in trace_ids:\n- trace_manager.register_trace(trace_id, create_test_trace_info(f\"tr-{trace_id}\", \"test\"))\n-\n- def add_spans(thread_id):\n- for trace_id in trace_ids:\n- trace_manager.register_span(\n- _create_test_span(f\"tr-{trace_id}\", trace_id=trace_id, span_id=thread_id)\n- )\n-\n- threads = [Thread(target=add_spans, args=[i]) for i in range(num_threads)]\n-\n- for thread in threads:\n- thread.start()\n- for thread in threads:\n- thread.join()\n-\n- for trace_id in trace_ids:\n- manager_trace = trace_manager.pop_trace(trace_id)\n- assert manager_trace is not None\n- assert manager_trace.trace.info.request_id == f\"tr-{trace_id}\"\n- assert len(manager_trace.trace.data.spans) == num_threads\n-\n-\n-def test_traces_buffer_expires_after_ttl(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_TRACE_BUFFER_TTL_SECONDS\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- time.sleep(1)\n-\n- assert request_id_1 not in trace_manager._traces\n-\n- # Clear singleton instance again to avoid side effects to other tests\n- InMemoryTraceManager._instance = None\n-\n-\n-def test_traces_buffer_expires_and_log_when_timeout_is_set(monkeypatch):\n- # Setting MLFLOW_TRACE_TIMEOUT_SECONDS let MLflow to periodically check the\n- # expired traces and log expired ones to the backend.\n- monkeypatch.setenv(\"MLFLOW_TRACE_TIMEOUT_SECONDS\", \"1\")\n- monkeypatch.setenv(\"MLFLOW_TRACE_TTL_CHECK_INTERVAL_SECONDS\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_info = create_test_trace_info(request_id_1, \"test\")\n- trace_manager.register_trace(12345, trace_info)\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert trace_manager._traces.get(request_id_1) is not None\n- assert len(trace_manager._traces[request_id_1].span_dict) == 1\n-\n- assert request_id_1 in trace_manager._traces\n-\n- time.sleep(3)\n-\n- assert request_id_1 not in trace_manager._traces\n- assert span_1_1.status.status_code == SpanStatusCode.ERROR\n-\n-\n-def test_traces_buffer_max_size_limit(monkeypatch):\n- monkeypatch.setenv(\"MLFLOW_TRACE_BUFFER_MAX_SIZE\", \"1\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"experiment\"))\n-\n- span_1_1 = _create_test_span(request_id_1)\n- trace_manager.register_span(span_1_1)\n-\n- assert request_id_1 in trace_manager._traces\n- assert len(trace_manager._traces) == 1\n-\n- request_id_2 = \"tr-2\"\n- trace_manager.register_trace(67890, create_test_trace_info(request_id_2, \"experiment\"))\n- span_2_1 = _create_test_span(request_id_2)\n- trace_manager.register_span(span_2_1)\n-\n- assert request_id_1 not in trace_manager._traces\n- assert request_id_2 in trace_manager._traces\n- assert len(trace_manager._traces) == 1\n-\n- # Clear singleton instance again to avoid side effects to other tests\n- InMemoryTraceManager._instance = None\n-\n-\n-def test_get_span_from_id():\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- span_1_1 = _create_test_span(request_id_1, trace_id=111, span_id=1)\n- span_1_2 = _create_test_span(request_id_1, trace_id=111, span_id=2, parent_id=1)\n-\n- request_id_2 = \"tr-2\"\n- trace_manager.register_trace(67890, create_test_trace_info(request_id_2, \"test\"))\n- span_2_1 = _create_test_span(request_id_2, trace_id=222, span_id=1)\n- span_2_2 = _create_test_span(request_id_2, trace_id=222, span_id=2, parent_id=1)\n-\n- # Add a span for a new trace\n- trace_manager.register_span(span_1_1)\n- trace_manager.register_span(span_1_2)\n- trace_manager.register_span(span_2_1)\n- trace_manager.register_span(span_2_2)\n-\n- assert trace_manager.get_span_from_id(request_id_1, span_1_1.span_id) == span_1_1\n- assert trace_manager.get_span_from_id(request_id_2, span_2_2.span_id) == span_2_2\n-\n-\n-def test_get_root_span_id():\n- trace_manager = InMemoryTraceManager.get_instance()\n-\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- span_1_1 = _create_test_span(request_id_1, span_id=1)\n- span_1_2 = _create_test_span(request_id_1, span_id=2, parent_id=1)\n-\n- # Add a span for a new trace\n- trace_manager.register_span(span_1_1)\n- trace_manager.register_span(span_1_2)\n-\n- assert trace_manager.get_root_span_id(request_id_1) == span_1_1.span_id\n-\n- # Non-existing trace\n- assert trace_manager.get_root_span_id(\"tr-2\") is None\n-\n-\n-def test_register_prompt():\n- prompt_version = PromptVersion(name=\"test_prompt\", version=1, template=\"Test template\")\n-\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- trace_manager.register_prompt(request_id_1, prompt_version)\n-\n- trace = trace_manager._traces[request_id_1]\n- assert trace.prompts == [prompt_version]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"}\n- ]\n-\n- # Adding multiple prompts to the same trace\n- prompt_version_2 = PromptVersion(name=\"test_prompt_2\", version=2, template=\"Test template 2\")\n-\n- trace_manager.register_prompt(request_id_1, prompt_version_2)\n- assert trace.prompts == [prompt_version, prompt_version_2]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"},\n- {\"name\": \"test_prompt_2\", \"version\": \"2\"},\n- ]\n-\n- # Registering the same prompt should not add it again\n- trace_manager.register_prompt(request_id_1, prompt_version)\n- assert trace.prompts == [prompt_version, prompt_version_2]\n- assert json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY]) == [\n- {\"name\": \"test_prompt\", \"version\": \"1\"},\n- {\"name\": \"test_prompt_2\", \"version\": \"2\"},\n- ]\n-\n-\n-def test_register_prompt_thread_safety():\n- trace_manager = InMemoryTraceManager.get_instance()\n- request_id_1 = \"tr-1\"\n- trace_manager.register_trace(12345, create_test_trace_info(request_id_1, \"test\"))\n- prompt_versions = [\n- PromptVersion(name=f\"test_prompt_{i}\", version=i, template=f\"Test template {i}\")\n- for i in range(10)\n- ]\n-\n- def register_prompt(prompt_version):\n- trace_manager.register_prompt(request_id_1, prompt_version)\n-\n- threads = [Thread(target=register_prompt, args=[pv]) for pv in prompt_versions]\n- for thread in threads:\n- thread.start()\n- for thread in threads:\n- thread.join()\n-\n- trace = trace_manager._traces[request_id_1]\n- assert len(trace.prompts) == len(prompt_versions)\n- assert len(json.loads(trace.info.tags[LINKED_PROMPTS_TAG_KEY])) == len(prompt_versions)\n-\n-\n-def _create_test_span(\n- request_id=\"tr-12345\",\n- trace_id: int = 12345,\n- span_id: int = 123,\n- parent_id: int | None = None,\n- start_time=None,\n- end_time=None,\n-):\n- mock_otel_span = create_mock_otel_span(\n- trace_id=trace_id,\n- span_id=span_id,\n- parent_id=parent_id,\n- start_time=start_time,\n- end_time=end_time,\n- )\n-\n- span = LiveSpan(mock_otel_span, request_id)\n- span.set_status(\"OK\")\n- return span\n"},"proof":{"kind":"string","value":"{\n \"case_contract\": \"Every case is a dict with exactly two keys:\\n\\n- `args`: a tuple or list of positional arguments for the target callable.\\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\\n callable.\\n\\n`check_case(case)` must invoke the target using\\n`target(*case[\\\"args\\\"], **case[\\\"kwargs\\\"])`. The evaluator may provide unseen\\ncases using the same representation. Argument values are constrained only by\\nthe natural-language specification and the target's public calling interface.\\nThe target must be the actual implementation callable, not a wrapper invented\\nby the witness or certificate. If an instance method is taken from its class,\\n`args` includes the receiver instance first; a method already taken from an\\ninstance is bound and does not need that extra argument.\\nCases may exercise normal, boundary, or error-handling behavior when that\\nbehavior is covered by the specification or public interface. Whether an\\nexception or any other observed behavior is a violation depends on what the\\nspecification requires for that case.\\nCases contain stimulus only and must not contain expected outcomes, expected\\nexceptions, failure flags, or mutant-specific information.\"\n}"},"metadata":{"kind":"string","value":"{\n \"preservation_probes\": [\n {\n \"spec_id\": \"S21\",\n \"shared_functions\": 1,\n \"shared_lines\": 0\n }\n ]\n}"}}}],"truncated":false,"partial":false},"paginationData":{"pageIndex":0,"numItemsPerPage":100,"numTotalItems":291,"offset":0,"length":100}},"jwt":"eyJhbGciOiJFZERTQSIsImtpZCI6IjVHZDBvd0g5MTM2eDZjc1FvbE1zcktNWUZoRVFoUm5PVVVybEpjOUhaUEEifQ.eyJyZWFkIjp0cnVlLCJwZXJtaXNzaW9ucyI6eyJyZXBvLmNvbnRlbnQucmVhZCI6dHJ1ZX0sInJlcG9JZHMiOlsiNmFjMzU3ZmMxYTc3OWIwYzYzYjlkZTgzIl0sImlhdCI6MTc5NDQ4NjU5OSwianRpIjoiYzJlNDI5MmMtN2E5ZS00YWQxLThkN2QtN2Y5YmE3MTdiYzdlIiwic3ViIjoiL2RhdGFzZXRzL2NveGhpbi9TcGVjR3VhcmRCZW5jaCIsImV4cCI6MTc5NDQ5MDE5OSwiaXNzIjoiaHR0cHM6Ly9odWdnaW5nZmFjZS5jbyJ9.0UL-7GGs4-tDO9WCQpqzSrRl2YLzD0yx7tLtMqgHUM4Dq0xzXUPC1w4inU4YoyMnmzQIELoH45vHUFuCJAFjAA","displayUrls":true,"splitSizeSummaries":[{"config":"default","split":"full","numRows":291,"numBytesParquet":1190739},{"config":"default","split":"lite","numRows":100,"numBytesParquet":1126732}]},"discussionsStats":{"closed":0,"open":0,"total":0},"fullWidth":true,"hasGatedAccess":true,"hasFullAccess":true,"isEmbedded":false,"savedQueries":{"community":[],"user":[]}}">
schema_version int64 3 3 | instance_id stringclasses 30
values | spec dict | mask_patch stringclasses 30
values | gold_patch stringclasses 30
values | mutant_patch stringlengths 295 1.33k | test_patch stringclasses 30
values | proof dict | metadata dict |
|---|---|---|---|---|---|---|---|---|
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S5",
"text": "The method resets: - Import tracking sets for regular and typing imports",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "335-339",
"function": "_reset"
}
]
} | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..884e1de8 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -333,11 +333,11 @@ class StubGenerator:
# but simplifies code quite a bit.
# Impor... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S6",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S7",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": ... |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S6",
"text": "The method resets: - Sub-module import tracking",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "337-337",
"function": "_reset"
}
]
} | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..919edbd3 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -335,7 +335,7 @@ class StubGenerator:
# Imports that are needed at the top of the file
... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S5",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S7",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": ... |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S7",
"text": "The method resets: - TypeVar definitions dictionary",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "341-341",
"function": "_reset"
}
]
} | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..2873d0e5 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -339,7 +339,8 @@ class StubGenerator:
# Typing imports (behind if TYPE_CHECKING) that are nee... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S5",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S6",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S9",
"text": "The method resets: - Generated stub content list",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "346-346",
"function": "_reset"
}
]
} | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..81c01813 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -344,7 +344,8 @@ class StubGenerator:
self._current_objects = {} # type: Dict[str, Any]
... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S5",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S6",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S10",
"text": "The method resets: - Current parent module reference",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "352-352",
"function": "_reset"
}
]
} | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..a9b4d539 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -349,7 +349,9 @@ class StubGenerator:
# These have a shorter "scope"
# Current paren... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S5",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S6",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S7",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "... |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S20",
"text": "name : str — The name of the class for which to generate the stub. clazz : type — The class object to analyze and generate a stub for",
"category": "structural",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "637-637",
"function... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..a5cdcb09 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -634,7 +634,8 @@ class StubGenerator:
)
return annotation_string
- def _gen... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S21",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S22",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S23",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id"... |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S24",
"text": "Processes static methods, class methods, and regular methods with appropriate decorators / Support decorator analysis and signature extraction",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "690-700",
... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..20922189 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -694,7 +694,7 @@ class StubGenerator:
for key, element in clazz.__dict__.items():
... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S23",
"shared_functions": 1,
"shared_lines": 11
},
{
"spec_id": "S27",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S20",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S27",
"text": "Skips __init__ method generation for TriggeredRun and DeployedFlow subclasses",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "639-639",
"function": "_generate_class_stub"
},
{
"file": "me... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..ecd199ce 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -636,7 +636,7 @@ class StubGenerator:
def _generate_class_stub(self, name: str, clazz: type) ->... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S23",
"shared_functions": 1,
"shared_lines": 5
},
{
"spec_id": "S24",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S20",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S32",
"text": "name : str — The name of the function for which to generate the stub. func : Optional[Union[Callable, classmethod]], optional — The actual function object to analyze. sign : Optional[List[inspect.Signature]], optional — List of function signatures to use instead of inspecting 'func'.",
... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..1e9f146d 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -1179,7 +1179,7 @@ class StubGenerator:
self,
name: str,
func: Optional[Uni... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S33",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S34",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S35",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id"... |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S33",
"text": "Either 'func' or 'sign' must be provided, but not necessarily both / func : ... If None, the 'sign' parameter must be provided / Exceptions: RuntimeError — Raised when neither 'func' nor 'sign' parameters are provided",
"category": "behavioral",
"code_mapping": [
{
"file":... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..a49a2494 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -1238,7 +1238,7 @@ class StubGenerator:
buff = StringIO()
if sign is None and func... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S32",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S34",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S35",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id"... |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S37",
"text": "Exceptions: ValueError — May be raised when inspect.signature() fails on the provided function (handled gracefully)",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "1248-1255",
"function": "_generate_fu... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..7faca31b 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -1244,7 +1244,7 @@ class StubGenerator:
)
try:
sign = sign or [insp... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S32",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S33",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S34",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S39",
"text": "Functions with STUBGEN_IGNORE docstring are ignored and return empty string",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "1256-1260",
"function": "_generate_function_stub"
}
]
} | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..98f564d0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -1253,7 +1253,7 @@ class StubGenerator:
if doc == "STUBGEN_IGNORE":
# Ignore met... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S41",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S32",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S33",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S41",
"text": "doc : Optional[str], optional — Custom documentation string to use. If None, uses func.__doc__",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "1256-1256",
"function": "_generate_function_stub"
},
... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..56ea64f3 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -1249,7 +1249,10 @@ class StubGenerator:
# which doesn't have a signature. We ignore for ... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 11
},
{
"spec_id": "S42",
"shared_functions": 1,
"shared_lines": 11
},
{
"spec_id": "S39",
"shared_functions": 1,
"shared_lines": 1
}
]
} |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S43",
"text": "deco : Optional[str], optional — Decorator string to apply to the function (e.g., \"@property\", \"@staticmethod\")",
"category": "behavioral",
"code_mapping": [
{
"file": "metaflow/cmd/develop/stub_generator.py",
"lines": "1281-1284",
"function": "_generate_fu... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..92b04dad 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -1271,7 +1271,7 @@ class StubGenerator:
# some IDEs seem to prefer if there is one no... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 4
},
{
"spec_id": "S42",
"shared_functions": 1,
"shared_lines": 4
},
{
"spec_id": "S34",
"shared_functions": 1,
"shared_lines": 2
}
]
} |
3 | Netflix__metaflow.b390a8d4.test_stub_generator.7bf08c98.lv1 | {
"spec_id": "S42",
"text": "Generate a stub definition for a Python function with proper type annotations and documentation. This method creates a stub representation of a function that can be used in .pyi files for type checking and IDE support. It handles function signatures, parameter defaults, type annotations... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
- def _reset(self):
- # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..9ee2a2e0 100644
+++ a/metaflow/cmd/develop/stub_generator.py
--- b/metaflow/cmd/develop/stub_generator.py
@@ -328,28 +328,28 @@ class StubGenerator:
self._reset()
+ def _reset(self):
+ # "... | diff --git a/metaflow/cmd/develop/stub_generator.py b/metaflow/cmd/develop/stub_generator.py
index 50d9ca29..cc73bbef 100644
--- a/metaflow/cmd/develop/stub_generator.py
+++ b/metaflow/cmd/develop/stub_generator.py
@@ -1325,7 +1325,8 @@ class StubGenerator:
init_blank = False
b... | diff --git a/test/cmd/develop/test_stub_generator.py b/test/cmd/develop/test_stub_generator.py
deleted file mode 100644
index c115c463..00000000
--- a/test/cmd/develop/test_stub_generator.py
+++ /dev/null
@@ -1,525 +0,0 @@
-import tempfile
-import typing
-import inspect
-from typing import TypeVar, Optional
-from unitt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 58
},
{
"spec_id": "S40",
"shared_functions": 1,
"shared_lines": 34
},
{
"spec_id": "S36",
"shared_functions": 1,
"shared_lines": 16
},
{
"spec_... |
3 | astropy__astropy.b0db0daa.test_comparison.445d81a3.lv1 | {
"spec_id": "S2",
"text": "The name field automatically converts values to strings. Sets _name attribute to None if value is None, otherwise converts to str.",
"category": "behavioral",
"code_mapping": [
{
"file": "/testbed/astropy/cosmology/_src/core.py",
"lines": "105-106",
"function": ... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
--- a/astropy/cosmology/__init__.py
+++ b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
-from ._src.funcs import cosmology_equal, z_at_value
+from ._src.f... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
+++ a/astropy/cosmology/__init__.py
--- b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
+from ._src.funcs import cosmology_equal, z_at_value
-from ._src.f... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..6f47b5dacd 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -103,7 +103,7 @@ class _NameField:
return instance._name
def __set__(self, instance: "Cosmology", value: str | Non... | diff --git a/astropy/cosmology/_src/tests/funcs/test_comparison.py b/astropy/cosmology/_src/tests/funcs/test_comparison.py
deleted file mode 100644
index 80999a7bc8..0000000000
--- a/astropy/cosmology/_src/tests/funcs/test_comparison.py
+++ /dev/null
@@ -1,337 +0,0 @@
-# Licensed under a 3-clause BSD style license - se... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S3",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S4",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_comparison.445d81a3.lv1 | {
"spec_id": "S3",
"text": "Check equality between Cosmologies.\n\nChecks the Parameters and immutable fields (i.e. not \"meta\").\n\nParameters\n----------\nother : `~astropy.cosmology.Cosmology` subclass instance, positional-only\n The object in which to compare.\n\nReturns\n-------\nbool\n `True` if Parame... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
--- a/astropy/cosmology/__init__.py
+++ b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
-from ._src.funcs import cosmology_equal, z_at_value
+from ._src.f... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
+++ a/astropy/cosmology/__init__.py
--- b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
+from ._src.funcs import cosmology_equal, z_at_value
-from ._src.f... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..3c623771ea 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -418,14 +418,12 @@ class Cosmology(metaclass=ABCMeta):
return NotImplemented # allows other.__eq__
return... | diff --git a/astropy/cosmology/_src/tests/funcs/test_comparison.py b/astropy/cosmology/_src/tests/funcs/test_comparison.py
deleted file mode 100644
index 80999a7bc8..0000000000
--- a/astropy/cosmology/_src/tests/funcs/test_comparison.py
+++ /dev/null
@@ -1,337 +0,0 @@
-# Licensed under a 3-clause BSD style license - se... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S4",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S2",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_comparison.445d81a3.lv1 | {
"spec_id": "S4",
"text": "If other.__class__ is not self.__class__, return NotImplemented to allow other.__eq__ to be called.",
"category": "behavioral",
"code_mapping": [
{
"file": "/testbed/astropy/cosmology/_src/core.py",
"lines": "419-420",
"function": "Cosmology.__eq__"
}
]
} | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
--- a/astropy/cosmology/__init__.py
+++ b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
-from ._src.funcs import cosmology_equal, z_at_value
+from ._src.f... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
+++ a/astropy/cosmology/__init__.py
--- b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
+from ._src.funcs import cosmology_equal, z_at_value
-from ._src.f... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..b98e281c46 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -415,7 +415,7 @@ class Cosmology(metaclass=ABCMeta):
`True` if Parameters and names are the same, `False` otherwise.
... | diff --git a/astropy/cosmology/_src/tests/funcs/test_comparison.py b/astropy/cosmology/_src/tests/funcs/test_comparison.py
deleted file mode 100644
index 80999a7bc8..0000000000
--- a/astropy/cosmology/_src/tests/funcs/test_comparison.py
+++ /dev/null
@@ -1,337 +0,0 @@
-# Licensed under a 3-clause BSD style license - se... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S3",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S2",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_comparison.445d81a3.lv1 | {
"spec_id": "S9",
"text": "The function internally calls `cosmology_equal` and returns its logical negation, providing a consistent way to check for cosmology inequality with the same parameter parsing and conversion capabilities.",
"category": "implementation_constraint",
"code_mapping": [
{
"file":... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
--- a/astropy/cosmology/__init__.py
+++ b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
-from ._src.funcs import cosmology_equal, z_at_value
+from ._src.f... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
+++ a/astropy/cosmology/__init__.py
--- b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
+from ._src.funcs import cosmology_equal, z_at_value
-from ._src.f... | diff --git a/astropy/cosmology/_src/funcs/comparison.py b/astropy/cosmology/_src/funcs/comparison.py
index 9225fdcdc9..d7285fc625 100644
--- a/astropy/cosmology/_src/funcs/comparison.py
+++ b/astropy/cosmology/_src/funcs/comparison.py
@@ -360,6 +360,13 @@ def _cosmology_not_equal(
astropy.cosmology.cosmology_equal... | diff --git a/astropy/cosmology/_src/tests/funcs/test_comparison.py b/astropy/cosmology/_src/tests/funcs/test_comparison.py
deleted file mode 100644
index 80999a7bc8..0000000000
--- a/astropy/cosmology/_src/tests/funcs/test_comparison.py
+++ /dev/null
@@ -1,337 +0,0 @@
-# Licensed under a 3-clause BSD style license - se... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S8",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S12",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_comparison.445d81a3.lv1 | {
"spec_id": "S10",
"text": "Parse Cosmology-like objects to |Cosmology| instances, using provided formats.\n\n``format`` is broadcast to match the shape of the cosmology arguments. Note\nthat the cosmology arguments are not broadcast against ``format``, so it\ncannot determine the output shape.\n\nParameters\n----... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
--- a/astropy/cosmology/__init__.py
+++ b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
-from ._src.funcs import cosmology_equal, z_at_value
+from ._src.f... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
+++ a/astropy/cosmology/__init__.py
--- b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
+from ._src.funcs import cosmology_equal, z_at_value
-from ._src.f... | diff --git a/astropy/cosmology/_src/funcs/comparison.py b/astropy/cosmology/_src/funcs/comparison.py
index 9225fdcdc9..a4c8ed0e3d 100644
--- a/astropy/cosmology/_src/funcs/comparison.py
+++ b/astropy/cosmology/_src/funcs/comparison.py
@@ -149,7 +149,8 @@ def _parse_formats(*cosmos: object, format: _FormatsType) -> ndar... | diff --git a/astropy/cosmology/_src/tests/funcs/test_comparison.py b/astropy/cosmology/_src/tests/funcs/test_comparison.py
deleted file mode 100644
index 80999a7bc8..0000000000
--- a/astropy/cosmology/_src/tests/funcs/test_comparison.py
+++ /dev/null
@@ -1,337 +0,0 @@
-# Licensed under a 3-clause BSD style license - se... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S12",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S8",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S9",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_comparison.445d81a3.lv1 | {
"spec_id": "S12",
"text": "Return element-wise equality check on the cosmologies.\n\n.. note::\n\n Cosmologies are currently scalar in their parameters.\n\nParameters\n----------\ncosmo1, cosmo2 : |Cosmology|-like\n The objects to compare. Must be convertible to |Cosmology|, as specified\n by ``format``.... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
--- a/astropy/cosmology/__init__.py
+++ b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
-from ._src.funcs import cosmology_equal, z_at_value
+from ._src.f... | diff --git a/astropy/cosmology/__init__.py b/astropy/cosmology/__init__.py
index 6f9df74213..13c1c03096 100644
+++ a/astropy/cosmology/__init__.py
--- b/astropy/cosmology/__init__.py
@@ -23,7 +23,7 @@ from ._src.flrw import (
wCDM,
wpwaCDM,
)
+from ._src.funcs import cosmology_equal, z_at_value
-from ._src.f... | diff --git a/astropy/cosmology/_src/funcs/comparison.py b/astropy/cosmology/_src/funcs/comparison.py
index 9225fdcdc9..d17bdc1331 100644
--- a/astropy/cosmology/_src/funcs/comparison.py
+++ b/astropy/cosmology/_src/funcs/comparison.py
@@ -312,7 +312,7 @@ def cosmology_equal(
if eq is NotImplemented:
... | diff --git a/astropy/cosmology/_src/tests/funcs/test_comparison.py b/astropy/cosmology/_src/tests/funcs/test_comparison.py
deleted file mode 100644
index 80999a7bc8..0000000000
--- a/astropy/cosmology/_src/tests/funcs/test_comparison.py
+++ /dev/null
@@ -1,337 +0,0 @@
-# Licensed under a 3-clause BSD style license - se... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S8",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S9",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_compression_failures.48dc420b.lv1 | {
"spec_id": "S1",
"text": "Check if a keyword exists in the header.\n\nThis method implements the 'in' operator for Header objects, allowing you to check whether a specific keyword is present in the header using syntax like `keyword in header`.\n\nParameters: keyword (str, int, or tuple) - The keyword to search fo... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..360cb90faf 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -144,7 +144,7 @@ class Header:
return True
try:
self._cardindex(keyword)
- except (KeyError, IndexError):
+ ... | diff --git a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py b/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
deleted file mode 100644
index 048bcb2232..0000000000
--- a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
+++ /dev/null
@@ -1,173 +0,0 @@
-# Licensed und... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S11",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S2",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id":... |
3 | astropy__astropy.b0db0daa.test_compression_failures.48dc420b.lv1 | {
"spec_id": "S2",
"text": "Returns an index into the ._cards list given a valid lookup key.\n\nThis method handles various types of keys to locate cards within the header's internal card list. It supports string keywords, integer indices, tuples for duplicate keyword access, and slices.\n\nParameters: key (str, in... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..a181b1738e 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1689,7 +1689,7 @@ class Header:
return key
elif isinstance(key, tuple):
if (
- len(key) != 2
+ ... | diff --git a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py b/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
deleted file mode 100644
index 048bcb2232..0000000000
--- a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
+++ /dev/null
@@ -1,173 +0,0 @@
-# Licensed und... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S11",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_compression_failures.48dc420b.lv1 | {
"spec_id": "S4",
"text": "Insert a new card before or after an existing card in the header.\n\nThis is an internal method used to implement support for the legacy before/after keyword arguments to Header.update() and related methods. It provides functionality to insert cards at specific positions relative to exis... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..9ef360aa17 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1786,8 +1786,6 @@ class Header:
if before is not None:
if old_idx == insertion_idx - 1:
return
- ... | diff --git a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py b/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
deleted file mode 100644
index 048bcb2232..0000000000
--- a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
+++ /dev/null
@@ -1,173 +0,0 @@
-# Licensed und... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S11",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_compression_failures.48dc420b.lv1 | {
"spec_id": "S8",
"text": "Similar to :meth:`dict.get`--returns the value associated with keyword in the header, or a default value if the keyword is not found.\n\nParameters: key (str) - A keyword that may or may not be in the header. default (optional) - A default value to return if the keyword is not found in t... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..457cd72876 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -885,7 +885,10 @@ class Header:
if the keyword is not in the header.
"""
try:
- return self[key]
+ ... | diff --git a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py b/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
deleted file mode 100644
index 048bcb2232..0000000000
--- a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
+++ /dev/null
@@ -1,173 +0,0 @@
-# Licensed und... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S11",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_compression_failures.48dc420b.lv1 | {
"spec_id": "S11",
"text": "Descriptor method to lazily load and return the Header object from stored header string data.\n\nThis method implements the descriptor protocol's __get__ method to provide lazy loading of Header objects. When an HDU's header is first accessed, this method converts the raw header string ... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..075f5687ab 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1929,7 +1929,7 @@ class _DelayedHeader:
except KeyError:
if obj._header_str is not None:
hdr = Header.fromstri... | diff --git a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py b/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
deleted file mode 100644
index 048bcb2232..0000000000
--- a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
+++ /dev/null
@@ -1,173 +0,0 @@
-# Licensed und... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S2",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": ... |
3 | astropy__astropy.b0db0daa.test_compression_failures.48dc420b.lv1 | {
"spec_id": "S14",
"text": "Parse TFORMn keyword for a binary table into a (repeat, format, option) tuple.\n\nThis function parses the FITS binary table column format specification (TFORMn) and extracts the repeat count, format code, and optional parameters.\n\nParameters: tform (str) - The TFORMn keyword value fr... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/column.py b/astropy/io/fits/column.py
index 804adb07c9..a7f15cbe93 100644
--- a/astropy/io/fits/column.py
+++ b/astropy/io/fits/column.py
@@ -2297,7 +2297,7 @@ def _parse_tformat(tform):
raise VerifyError(f"Format {tform!r} is not recognized.")
if repeat == "":
- repe... | diff --git a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py b/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
deleted file mode 100644
index 048bcb2232..0000000000
--- a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
+++ /dev/null
@@ -1,173 +0,0 @@
-# Licensed und... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S15",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_compression_failures.48dc420b.lv1 | {
"spec_id": "S15",
"text": "Parse the TFORMn keywords for ASCII tables into a (format, width, precision) tuple.\n\nThis function parses ASCII table column format specifications from FITS headers. The precision value is always zero unless the format is one of 'E', 'F', or 'D' (floating-point formats).\n\nParameters... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..6d62e97c84 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/column.py b/astropy/io/fits/column.py
index 804adb07c9..d4544b7ce1 100644
--- a/astropy/io/fits/column.py
+++ b/astropy/io/fits/column.py
@@ -2340,7 +2340,7 @@ def _parse_ascii_tformat(tform, strict=False):
else:
# Just use a default width of 0 if unspecified
... | diff --git a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py b/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
deleted file mode 100644
index 048bcb2232..0000000000
--- a/astropy/io/fits/hdu/compressed/tests/test_compression_failures.py
+++ /dev/null
@@ -1,173 +0,0 @@
-# Licensed und... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S14",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S1",
"text": "Return the state of the unit for pickling. This method is called during the pickling process to get the object's state that should be serialized. Returns dict: A dictionary containing the object's state.",
"category": "behavioral",
"code_mapping": [
{
"file": "astropy/units... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/units/core.py b/astropy/units/core.py
index 1fa99c38e1..7d260596ed 100644
--- a/astropy/units/core.py
+++ b/astropy/units/core.py
@@ -374,6 +374,7 @@ class UnitBase:
state = self.__dict__.copy()
state.pop("_hash", None)
state.pop("_physical_type_id", None)
+ state.... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S2",
"shared_functions": 1,
"shared_lines": 3
},
{
"spec_id": "S3",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S4",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "... |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S2",
"text": "It removes cached properties that depend on string hashes, which can vary between Python sessions, to ensure consistent behavior when unpickling units in different sessions. Specifically, the '_hash' and '_physical_type_id' cached properties are excluded from the returned state since the... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/units/core.py b/astropy/units/core.py
index 1fa99c38e1..6c3aafaafe 100644
--- a/astropy/units/core.py
+++ b/astropy/units/core.py
@@ -373,7 +373,6 @@ class UnitBase:
# hashes of strings vary between sessions.
state = self.__dict__.copy()
state.pop("_hash", None)
- ... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 1,
"shared_lines": 3
},
{
"spec_id": "S3",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S4",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "... |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S3",
"text": "Enter the unit context manager. This method is called when entering a `with` statement block that uses a `_UnitContext` instance as a context manager. Returns None: This method does not return any value. The actual unit registry modification occurs during the `_UnitContext` initializatio... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/units/core.py b/astropy/units/core.py
index 1fa99c38e1..a5651dabba 100644
--- a/astropy/units/core.py
+++ b/astropy/units/core.py
@@ -1448,7 +1448,7 @@ class _UnitContext:
_unit_registries.append(_UnitRegistry(init=init, equivalencies=equivalencies))
def __enter__(self) -> None:
- ... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S2",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S4",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S5",
"text": "def __exit__(self, type: type[BaseException] | None, value: BaseException | None, tb: TracebackType | None) -> None, where type is the exception type that caused the context to exit, or None if no exception occurred; value is the exception instance; tb is the traceback object. Returns No... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/units/core.py b/astropy/units/core.py
index 1fa99c38e1..ff66207d34 100644
--- a/astropy/units/core.py
+++ b/astropy/units/core.py
@@ -1457,6 +1457,8 @@ class _UnitContext:
tb: TracebackType | None,
) -> None:
_unit_registries.pop()
+ if type is not None:
+ r... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S4",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S2",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S14",
"text": "Converts string names to their corresponding cosmology realizations from the registry. If a string, it should be the name of a built-in cosmology realization (e.g., 'Planck18', 'WMAP7'). Returns astropy.cosmology.Cosmology: A validated Cosmology instance corresponding to the input value... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/default.py b/astropy/cosmology/_src/default.py
index 48ca309529..ed8788b60a 100644
--- a/astropy/cosmology/_src/default.py
+++ b/astropy/cosmology/_src/default.py
@@ -37,6 +37,8 @@ class default_cosmology(ScienceState):
_default_value: ClassVar[str] = "Planck18"
_value: C... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S13",
"shared_functions": 1,
"shared_lines": 4
},
{
"spec_id": "S15",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S16",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id"... |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S15",
"text": "Validates that Cosmology instances are passed through unchanged. If a Cosmology instance, it is returned unchanged.",
"category": "behavioral",
"code_mapping": [
{
"file": "astropy/cosmology/_src/default.py",
"lines": "66-71",
"function": "default_cosmology.val... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/default.py b/astropy/cosmology/_src/default.py
index 48ca309529..fa49d550ce 100644
--- a/astropy/cosmology/_src/default.py
+++ b/astropy/cosmology/_src/default.py
@@ -71,6 +71,13 @@ class default_cosmology(ScienceState):
"default_cosmology must be a string or Cosmolo... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S16",
"shared_functions": 1,
"shared_lines": 6
},
{
"spec_id": "S13",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S14",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id"... |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S16",
"text": "Raises TypeError: If value is not None, a string, or a Cosmology instance.",
"category": "behavioral",
"code_mapping": [
{
"file": "astropy/cosmology/_src/default.py",
"lines": "66-71",
"function": "default_cosmology.validate"
}
]
} | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/default.py b/astropy/cosmology/_src/default.py
index 48ca309529..b4f55d277d 100644
--- a/astropy/cosmology/_src/default.py
+++ b/astropy/cosmology/_src/default.py
@@ -66,7 +66,7 @@ class default_cosmology(ScienceState):
value = None
else:
... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S15",
"shared_functions": 1,
"shared_lines": 6
},
{
"spec_id": "S13",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S14",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S17",
"text": "Raises ValueError: If value is a string but does not correspond to any built-in cosmology realization available in the registry.",
"category": "behavioral",
"code_mapping": [
{
"file": "astropy/cosmology/_src/default.py",
"lines": "95-105",
"function": "default... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/default.py b/astropy/cosmology/_src/default.py
index 48ca309529..ad256bb084 100644
--- a/astropy/cosmology/_src/default.py
+++ b/astropy/cosmology/_src/default.py
@@ -97,6 +97,9 @@ class default_cosmology(ScienceState):
"""
from astropy.cosmology import realizations... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S13",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S14",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S15",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_realizations.5b9cf523.lv1 | {
"spec_id": "S18",
"text": "@classmethod def validate(cls, value: Cosmology | str | None) -> Cosmology | None, with parameter value : None, str, or astropy.cosmology.Cosmology, the cosmology specification to validate.",
"category": "structural",
"code_mapping": [
{
"file": "astropy/cosmology/_src/def... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
--- a/astropy/cosmology/_src/core.py
+++ b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
- def __get__(
- self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/core.py b/astropy/cosmology/_src/core.py
index 0d561c169c..d1073ddc2e 100644
+++ a/astropy/cosmology/_src/core.py
--- b/astropy/cosmology/_src/core.py
@@ -93,17 +93,17 @@ class _NameField:
default: str | None = None
+ def __get__(
+ self, instance: Union["Cosmology... | diff --git a/astropy/cosmology/_src/default.py b/astropy/cosmology/_src/default.py
index 48ca309529..7a8d2104dd 100644
--- a/astropy/cosmology/_src/default.py
+++ b/astropy/cosmology/_src/default.py
@@ -39,7 +39,7 @@ class default_cosmology(ScienceState):
_value: ClassVar[str | Cosmology] = "Planck18"
@clas... | diff --git a/astropy/cosmology/_src/tests/test_realizations.py b/astropy/cosmology/_src/tests/test_realizations.py
deleted file mode 100644
index bfe7a1352e..0000000000
--- a/astropy/cosmology/_src/tests/test_realizations.py
+++ /dev/null
@@ -1,105 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rs... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S13",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S14",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S15",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_table.48eef659.lv1 | {
"spec_id": "S25",
"text": "table : `astropy.table.Table`\n An Astropy Table instance containing the data and metadata from\n this VOTable. The table's meta dictionary will contain relevant\n table-level metadata such as ID, name, ref, ucd, utype, and\n description if present.",
"category": "behavior... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
--- a/astropy/io/votable/converters.py
+++ b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
-def _make_masked_array(data, mask):
- """
- Masked arrays of zero leng... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
+++ a/astropy/io/votable/converters.py
--- b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
+def _make_masked_array(data, mask):
+ """
+ Masked arrays of zero leng... | diff --git a/astropy/io/votable/tree.py b/astropy/io/votable/tree.py
index 31af4b7353..05e1f7e839 100644
--- a/astropy/io/votable/tree.py
+++ b/astropy/io/votable/tree.py
@@ -3516,7 +3516,7 @@ class TableElement(
from astropy.table import Table
meta = {}
- for key in ["ID", "name", "ref", "uc... | diff --git a/astropy/io/votable/tests/test_table.py b/astropy/io/votable/tests/test_table.py
deleted file mode 100644
index c01997ea9a..0000000000
--- a/astropy/io/votable/tests/test_table.py
+++ /dev/null
@@ -1,996 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rst
-"""
-Test the conversion to/fr... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S28",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S29",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_table.48eef659.lv1 | {
"spec_id": "S29",
"text": "Raises IndexError if no non-empty table is found in the VOTable file.",
"category": "behavioral",
"code_mapping": [
{
"file": "/testbed/astropy/io/votable/tree.py",
"lines": "4498-4505",
"function": "get_first_table"
}
]
} | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
--- a/astropy/io/votable/converters.py
+++ b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
-def _make_masked_array(data, mask):
- """
- Masked arrays of zero leng... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
+++ a/astropy/io/votable/converters.py
--- b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
+def _make_masked_array(data, mask):
+ """
+ Masked arrays of zero leng... | diff --git a/astropy/io/votable/tree.py b/astropy/io/votable/tree.py
index 31af4b7353..db77f86982 100644
--- a/astropy/io/votable/tree.py
+++ b/astropy/io/votable/tree.py
@@ -4503,7 +4503,7 @@ class VOTableFile(Element, _IDProperty, _DescriptionProperty):
for table in self.iter_tables():
if not ta... | diff --git a/astropy/io/votable/tests/test_table.py b/astropy/io/votable/tests/test_table.py
deleted file mode 100644
index c01997ea9a..0000000000
--- a/astropy/io/votable/tests/test_table.py
+++ /dev/null
@@ -1,996 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rst
-"""
-Test the conversion to/fr... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S28",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id": "S25",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_table.48eef659.lv1 | {
"spec_id": "S38",
"text": "Add a new column to the table using ``col`` as input. If ``index`` is supplied then insert column before ``index`` position in the list of columns, otherwise append column to the end of the list.\n\nThe ``col`` input can be any data object which is acceptable as a `~astropy.table.Table... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
--- a/astropy/io/votable/converters.py
+++ b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
-def _make_masked_array(data, mask):
- """
- Masked arrays of zero leng... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
+++ a/astropy/io/votable/converters.py
--- b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
+def _make_masked_array(data, mask):
+ """
+ Masked arrays of zero leng... | diff --git a/astropy/table/table.py b/astropy/table/table.py
index e08b790d23..10c1236c2f 100644
--- a/astropy/table/table.py
+++ b/astropy/table/table.py
@@ -2474,7 +2474,7 @@ class Table:
if index is not None:
# Move the other cols to the right of the new one
- move_names = self.col... | diff --git a/astropy/io/votable/tests/test_table.py b/astropy/io/votable/tests/test_table.py
deleted file mode 100644
index c01997ea9a..0000000000
--- a/astropy/io/votable/tests/test_table.py
+++ /dev/null
@@ -1,996 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rst
-"""
-Test the conversion to/fr... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S42",
"shared_functions": 1,
"shared_lines": 143
},
{
"spec_id": "S43",
"shared_functions": 1,
"shared_lines": 143
},
{
"spec_id": "S44",
"shared_functions": 1,
"shared_lines": 143
},
{
"sp... |
3 | astropy__astropy.b0db0daa.test_table.48eef659.lv1 | {
"spec_id": "S42",
"text": "rename_duplicate : bool\n Uniquify column name if it already exist. Default is False.",
"category": "structural",
"code_mapping": [
{
"file": "/testbed/astropy/table/table.py",
"lines": "2340-2482",
"function": "add_column"
}
]
} | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
--- a/astropy/io/votable/converters.py
+++ b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
-def _make_masked_array(data, mask):
- """
- Masked arrays of zero leng... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
+++ a/astropy/io/votable/converters.py
--- b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
+def _make_masked_array(data, mask):
+ """
+ Masked arrays of zero leng... | diff --git a/astropy/table/table.py b/astropy/table/table.py
index e08b790d23..a3285fe86d 100644
--- a/astropy/table/table.py
+++ b/astropy/table/table.py
@@ -2342,7 +2342,7 @@ class Table:
col,
index=None,
name=None,
- rename_duplicate=False,
+ rename_duplicate=True,
c... | diff --git a/astropy/io/votable/tests/test_table.py b/astropy/io/votable/tests/test_table.py
deleted file mode 100644
index c01997ea9a..0000000000
--- a/astropy/io/votable/tests/test_table.py
+++ /dev/null
@@ -1,996 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rst
-"""
-Test the conversion to/fr... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 143
},
{
"spec_id": "S43",
"shared_functions": 1,
"shared_lines": 143
},
{
"spec_id": "S44",
"shared_functions": 1,
"shared_lines": 143
}
]
} |
3 | astropy__astropy.b0db0daa.test_table.48eef659.lv1 | {
"spec_id": "S43",
"text": "copy : bool\n Make a copy of the new column. Default is True.",
"category": "structural",
"code_mapping": [
{
"file": "/testbed/astropy/table/table.py",
"lines": "2340-2482",
"function": "add_column"
}
]
} | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
--- a/astropy/io/votable/converters.py
+++ b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
-def _make_masked_array(data, mask):
- """
- Masked arrays of zero leng... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
+++ a/astropy/io/votable/converters.py
--- b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
+def _make_masked_array(data, mask):
+ """
+ Masked arrays of zero leng... | diff --git a/astropy/table/table.py b/astropy/table/table.py
index e08b790d23..31f3031a53 100644
--- a/astropy/table/table.py
+++ b/astropy/table/table.py
@@ -2430,7 +2430,7 @@ class Table:
# Note that along with the lines above and below, this allows broadcasting
# of scalars to the correct shape for... | diff --git a/astropy/io/votable/tests/test_table.py b/astropy/io/votable/tests/test_table.py
deleted file mode 100644
index c01997ea9a..0000000000
--- a/astropy/io/votable/tests/test_table.py
+++ /dev/null
@@ -1,996 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rst
-"""
-Test the conversion to/fr... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 143
},
{
"spec_id": "S42",
"shared_functions": 1,
"shared_lines": 143
},
{
"spec_id": "S44",
"shared_functions": 1,
"shared_lines": 143
}
]
} |
3 | astropy__astropy.b0db0daa.test_table.48eef659.lv1 | {
"spec_id": "S44",
"text": "default_name : str or None\n Name to use if both ``name`` and ``col.info.name`` are not available.\n Defaults to ``col{number_of_columns}``.",
"category": "structural",
"code_mapping": [
{
"file": "/testbed/astropy/table/table.py",
"lines": "2340-2482",
"... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
--- a/astropy/io/votable/converters.py
+++ b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
-def _make_masked_array(data, mask):
- """
- Masked arrays of zero leng... | diff --git a/astropy/io/votable/converters.py b/astropy/io/votable/converters.py
index 77184900ce..522e98ce05 100644
+++ a/astropy/io/votable/converters.py
--- b/astropy/io/votable/converters.py
@@ -72,85 +72,85 @@ else:
return x
+def _make_masked_array(data, mask):
+ """
+ Masked arrays of zero leng... | diff --git a/astropy/table/table.py b/astropy/table/table.py
index e08b790d23..e863930c3b 100644
--- a/astropy/table/table.py
+++ b/astropy/table/table.py
@@ -2424,7 +2424,7 @@ class Table:
2 0.2 b y
"""
if default_name is None:
- default_name = f"col{len(self.columns)}"... | diff --git a/astropy/io/votable/tests/test_table.py b/astropy/io/votable/tests/test_table.py
deleted file mode 100644
index c01997ea9a..0000000000
--- a/astropy/io/votable/tests/test_table.py
+++ /dev/null
@@ -1,996 +0,0 @@
-# Licensed under a 3-clause BSD style license - see LICENSE.rst
-"""
-Test the conversion to/fr... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 143
},
{
"spec_id": "S42",
"shared_functions": 1,
"shared_lines": 143
},
{
"spec_id": "S43",
"shared_functions": 1,
"shared_lines": 143
}
]
} |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S8",
"text": "The method performs an early return optimization if `idx` is greater than the number of cards in the header, as no indices would need updating in that case.",
"category": "implementation_constraint",
"code_mapping": [
{
"file": "/testbed/astropy/io/fits/header.py",
"l... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..ffae5e2fc8 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1807,10 +1807,6 @@ class Header:
For all cards with index above idx, increment or decrement its index
value in the keyword_indices ... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S11",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S12",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S11",
"text": "useblanks : bool, optional - If there are blank cards at the end of the Header, replace the first blank card so that the total number of cards in the Header does not increase. Otherwise preserve the number of blank cards. Default is True.",
"category": "structural",
"code_mapping": ... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..f31e7a10c0 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1167,7 +1167,7 @@ class Header:
self._keyword_indices[keyword].sort()
# Finally, if useblanks, delete a blank cards f... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S9",
"shared_functions": 1,
"shared_lines": 7
},
{
"spec_id": "S12",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S14",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id":... |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S14",
"text": "Raises ValueError if the card parameter is not a valid keyword string, tuple, Card object, or None.",
"category": "behavioral",
"code_mapping": [
{
"file": "/testbed/astropy/io/fits/header.py",
"lines": "1125-1135",
"function": "append"
}
]
} | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..320623f402 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1129,10 +1129,7 @@ class Header:
elif card is None:
card = Card()
elif not isinstance(card, Card):
- raise ... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S9",
"shared_functions": 1,
"shared_lines": 11
},
{
"spec_id": "S11",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S12",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S20",
"text": "keyword : str - The keyword to count instances of in the header",
"category": "structural",
"code_mapping": [
{
"file": "/testbed/astropy/io/fits/header.py",
"lines": "1288-1307",
"function": "count"
}
]
} | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..7f13f0ff58 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1286,7 +1286,7 @@ class Header:
for card in extend_cards:
self.append(card, useblanks=useblanks, bottom=bottom, end=end)
- ... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S19",
"shared_functions": 1,
"shared_lines": 20
},
{
"spec_id": "S24",
"shared_functions": 1,
"shared_lines": 13
},
{
"spec_id": "S22",
"shared_functions": 1,
"shared_lines": 3
},
{
"spec_i... |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S22",
"text": "Raises KeyError if the keyword is not found in the header",
"category": "behavioral",
"code_mapping": [
{
"file": "/testbed/astropy/io/fits/header.py",
"lines": "1300-1302",
"function": "count"
}
]
} | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..da52cedad7 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1302,7 +1302,7 @@ class Header:
# We have to look before we leap, since otherwise _keyword_indices,
# being a defaultdict, will cre... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S19",
"shared_functions": 1,
"shared_lines": 3
},
{
"spec_id": "S20",
"shared_functions": 1,
"shared_lines": 3
},
{
"spec_id": "S24",
"shared_functions": 1,
"shared_lines": 3
}
]
} |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S23",
"text": "The keyword lookup is case-insensitive and follows FITS keyword normalization rules.",
"category": "behavioral",
"code_mapping": [
{
"file": "/testbed/astropy/io/fits/header.py",
"lines": "1295-1296",
"function": "count"
}
]
} | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/header.py b/astropy/io/fits/header.py
index 0ca9aac9e8..9081ca2d6a 100644
--- a/astropy/io/fits/header.py
+++ b/astropy/io/fits/header.py
@@ -1297,7 +1297,8 @@ class Header:
The keyword to count instances of in the header
"""
- keyword = Card.normalize_keyword... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S19",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S20",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S24",
"shared_functions": 1,
"shared_lines": 2
}
]
} |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S27",
"text": "verbose : bool, optional - If True (default), enables verbose output including warnings about missing extensions, keywords, or other issues encountered during processing. If False, suppresses most warning messages.",
"category": "structural",
"code_mapping": [
{
"file": "/... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/scripts/fitsheader.py b/astropy/io/fits/scripts/fitsheader.py
index 1d6f9c7691..f4dcb611e7 100644
--- a/astropy/io/fits/scripts/fitsheader.py
+++ b/astropy/io/fits/scripts/fitsheader.py
@@ -108,7 +108,7 @@ class HeaderFormatter:
If `filename` does not exist or cannot be read.
... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S26",
"shared_functions": 1,
"shared_lines": 4
},
{
"spec_id": "S28",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S29",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id"... |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S29",
"text": "The constructor automatically opens the FITS file using astropy.io.fits.open().",
"category": "implementation_constraint",
"code_mapping": [
{
"file": "/testbed/astropy/io/fits/scripts/fitsheader.py",
"lines": "114",
"function": "__init__"
}
]
} | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/scripts/fitsheader.py b/astropy/io/fits/scripts/fitsheader.py
index 1d6f9c7691..6053a588c3 100644
--- a/astropy/io/fits/scripts/fitsheader.py
+++ b/astropy/io/fits/scripts/fitsheader.py
@@ -109,9 +109,10 @@ class HeaderFormatter:
"""
def __init__(self, filename, verbose=True):
+... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S26",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S27",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S28",
"shared_functions": 1,
"shared_lines": 1
}
]
} |
3 | astropy__astropy.b0db0daa.test_tilde_path.383d7c0c.lv1 | {
"spec_id": "S35",
"text": "The method internally calls the close() method on the astropy.io.fits.HDUList object that was created when the FITS file was opened.",
"category": "implementation_constraint",
"code_mapping": [
{
"file": "/testbed/astropy/io/fits/scripts/fitsheader.py",
"lines": "235... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
--- a/astropy/io/fits/card.py
+++ b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
-from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/card.py b/astropy/io/fits/card.py
index fac935694b..83d1f7a94b 100644
+++ a/astropy/io/fits/card.py
--- b/astropy/io/fits/card.py
@@ -8,7 +8,7 @@ import numpy as np
from astropy.utils.exceptions import AstropyUserWarning
from . import conf
+from .util import _is_int, _str_to_num, _words... | diff --git a/astropy/io/fits/scripts/fitsheader.py b/astropy/io/fits/scripts/fitsheader.py
index 1d6f9c7691..f4f8b9c381 100644
--- a/astropy/io/fits/scripts/fitsheader.py
+++ b/astropy/io/fits/scripts/fitsheader.py
@@ -232,7 +232,7 @@ class HeaderFormatter:
return cards
def close(self):
- self._h... | diff --git a/astropy/io/fits/tests/test_tilde_path.py b/astropy/io/fits/tests/test_tilde_path.py
deleted file mode 100644
index f5bfd0f7da..0000000000
--- a/astropy/io/fits/tests/test_tilde_path.py
+++ /dev/null
@@ -1,174 +0,0 @@
-# Licensed under a 3-clause BSD style license - see PYFITS.rst
-
-import os
-
-import num... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S26",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S27",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S28",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | matplotlib__matplotlib.86a476d2.test_backend_registry.872ba384.lv1 | {
"spec_id": "S38",
"text": "This function only loads entry points if they have not already been loaded and the backend is not built-in and not of ``module://some.backend`` format.",
"category": "behavioral",
"code_mapping": [
{
"file": "lib/matplotlib/backends/registry.py",
"lines": "398-405",
... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..55e4d8fbdf 100644
--- a/lib/matplotlib/backends/registry.py
+++ b/lib/matplotlib/backends/registry.py
@@ -110,94 +110,94 @@ class BackendRegistry:
# Clear all dynamically-added data, used for testing only.
... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..55e4d8fbdf 100644
+++ a/lib/matplotlib/backends/registry.py
--- b/lib/matplotlib/backends/registry.py
@@ -110,94 +110,94 @@ class BackendRegistry:
# Clear all dynamically-added data, used for testing only.
... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..70a6fb08a0 100644
--- a/lib/matplotlib/backends/registry.py
+++ b/lib/matplotlib/backends/registry.py
@@ -394,6 +394,7 @@ class BackendRegistry:
framework : str or None
The GUI framework, which... | diff --git a/lib/matplotlib/tests/test_backend_registry.py b/lib/matplotlib/tests/test_backend_registry.py
deleted file mode 100644
index 2bd8e161bd..0000000000
--- a/lib/matplotlib/tests/test_backend_registry.py
+++ /dev/null
@@ -1,169 +0,0 @@
-from collections.abc import Sequence
-from typing import Any
-
-import pyt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S37",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id": "S40",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id": "S42",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id"... |
3 | matplotlib__matplotlib.86a476d2.test_backend_registry.872ba384.lv1 | {
"spec_id": "S40",
"text": "Returns: backend : str - The backend name corresponding to the resolved GUI framework or backend. For GUI framework inputs, returns the preferred backend for that framework. For backend name inputs, returns the same backend name (potentially normalized). framework : str or None - The GU... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..55e4d8fbdf 100644
--- a/lib/matplotlib/backends/registry.py
+++ b/lib/matplotlib/backends/registry.py
@@ -110,94 +110,94 @@ class BackendRegistry:
# Clear all dynamically-added data, used for testing only.
... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..55e4d8fbdf 100644
+++ a/lib/matplotlib/backends/registry.py
--- b/lib/matplotlib/backends/registry.py
@@ -110,94 +110,94 @@ class BackendRegistry:
# Clear all dynamically-added data, used for testing only.
... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..b9308b72f9 100644
--- a/lib/matplotlib/backends/registry.py
+++ b/lib/matplotlib/backends/registry.py
@@ -400,6 +400,9 @@ class BackendRegistry:
# First check if it is a gui loop name.
backend = se... | diff --git a/lib/matplotlib/tests/test_backend_registry.py b/lib/matplotlib/tests/test_backend_registry.py
deleted file mode 100644
index 2bd8e161bd..0000000000
--- a/lib/matplotlib/tests/test_backend_registry.py
+++ /dev/null
@@ -1,169 +0,0 @@
-from collections.abc import Sequence
-from typing import Any
-
-import pyt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S37",
"shared_functions": 1,
"shared_lines": 9
},
{
"spec_id": "S42",
"shared_functions": 1,
"shared_lines": 9
},
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id"... |
3 | matplotlib__matplotlib.86a476d2.test_backend_registry.872ba384.lv1 | {
"spec_id": "S41",
"text": "Raises RuntimeError: If the input string is not recognized as either a valid GUI framework name or a valid backend name. The error message will indicate that the string is not a recognized GUI loop or backend name.",
"category": "behavioral",
"code_mapping": [
{
"file": "l... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..55e4d8fbdf 100644
--- a/lib/matplotlib/backends/registry.py
+++ b/lib/matplotlib/backends/registry.py
@@ -110,94 +110,94 @@ class BackendRegistry:
# Clear all dynamically-added data, used for testing only.
... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..55e4d8fbdf 100644
+++ a/lib/matplotlib/backends/registry.py
--- b/lib/matplotlib/backends/registry.py
@@ -110,94 +110,94 @@ class BackendRegistry:
# Clear all dynamically-added data, used for testing only.
... | diff --git a/lib/matplotlib/backends/registry.py b/lib/matplotlib/backends/registry.py
index 7f5a53f917..738dd928f0 100644
--- a/lib/matplotlib/backends/registry.py
+++ b/lib/matplotlib/backends/registry.py
@@ -406,6 +406,8 @@ class BackendRegistry:
try:
return self.resolve_backend(gui_or_backend)... | diff --git a/lib/matplotlib/tests/test_backend_registry.py b/lib/matplotlib/tests/test_backend_registry.py
deleted file mode 100644
index 2bd8e161bd..0000000000
--- a/lib/matplotlib/tests/test_backend_registry.py
+++ /dev/null
@@ -1,169 +0,0 @@
-from collections.abc import Sequence
-from typing import Any
-
-import pyt... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S37",
"shared_functions": 1,
"shared_lines": 4
},
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 4
},
{
"spec_id": "S40",
"shared_functions": 1,
"shared_lines": 4
}
]
} |
3 | mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1 | {
"spec_id": "S1",
"text": "Create a new webhook. This method creates a webhook that will be triggered when specified MLflow events occur. Webhooks allow you to receive HTTP notifications when events like model version creation, transition, or deletion happen in your MLflow deployment. Signature: create_webhook(sel... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py
index 8a34c23cd..203c145da 100644
--- a/mlflow/tracking/client.py
+++ b/mlflow/tracking/client.py
@@ -6337,6 +6337,9 @@ class MlflowClient:
if status is not None:
status = WebhookStatus(status) if isinstance(status, str) else statu... | diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py
deleted file mode 100644
index 97fd61b00..000000000
--- a/tests/tracking/test_client_webhooks.py
+++ /dev/null
@@ -1,219 +0,0 @@
-from pathlib import Path
-from typing import Iterator
-
-import pytest
-from cryptography.fernet ... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S3",
"shared_functions": 2,
"shared_lines": 28
},
{
"spec_id": "S4",
"shared_functions": 2,
"shared_lines": 5
},
{
"spec_id": "S2",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": ... |
3 | mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1 | {
"spec_id": "S4",
"text": "Raises: MlflowException: If webhook creation fails due to invalid parameters, network issues, or server-side errors. Common causes include invalid URL format, unsupported event types, or insufficient permissions. Task statement: Implement robust error handling for webhook delivery failur... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py
index 8a34c23cd..20dc8f033 100644
--- a/mlflow/tracking/client.py
+++ b/mlflow/tracking/client.py
@@ -6337,14 +6337,17 @@ class MlflowClient:
if status is not None:
status = WebhookStatus(status) if isinstance(status, str) else sta... | diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py
deleted file mode 100644
index 97fd61b00..000000000
--- a/tests/tracking/test_client_webhooks.py
+++ /dev/null
@@ -1,219 +0,0 @@
-from pathlib import Path
-from typing import Iterator
-
-import pytest
-from cryptography.fernet ... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 2,
"shared_lines": 5
},
{
"spec_id": "S3",
"shared_functions": 2,
"shared_lines": 1
},
{
"spec_id": "S2",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1 | {
"spec_id": "S6",
"text": "Raises: mlflow.MlflowException: If the webhook with the specified ID does not exist or if there are permission issues accessing the webhook.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/tracking/client.py",
"lines": "6366",
"function": "Mlflow... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py
index e1ef46902..a7e2ebea0 100644
--- a/mlflow/tracking/_model_registry/client.py
+++ b/mlflow/tracking/_model_registry/client.py
@@ -833,6 +833,8 @@ class ModelRegistryClient:
Returns:
A :py:class:`... | diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py
deleted file mode 100644
index 97fd61b00..000000000
--- a/tests/tracking/test_client_webhooks.py
+++ /dev/null
@@ -1,219 +0,0 @@
-from pathlib import Path
-from typing import Iterator
-
-import pytest
-from cryptography.fernet ... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S5",
"shared_functions": 2,
"shared_lines": 2
},
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1 | {
"spec_id": "S7",
"text": "List webhooks. Signature: list_webhooks(self, max_results: int | None = None, page_token: str | None = None) -> PagedList[Webhook]. Args: max_results: Maximum number of webhooks to return. page_token: Token specifying the next page of results. Returns: A :py:class:`mlflow.store.entities.... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py
index e1ef46902..dbb13d2e5 100644
--- a/mlflow/tracking/_model_registry/client.py
+++ b/mlflow/tracking/_model_registry/client.py
@@ -851,6 +851,9 @@ class ModelRegistryClient:
Returns:
A :py:class:`... | diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py
deleted file mode 100644
index 97fd61b00..000000000
--- a/tests/tracking/test_client_webhooks.py
+++ /dev/null
@@ -1,219 +0,0 @@
-from pathlib import Path
-from typing import Iterator
-
-import pytest
-from cryptography.fernet ... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S11",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1 | {
"spec_id": "S9",
"text": "Only the fields you specify will be updated; other fields will remain unchanged. name: If None, the current name is preserved. description: If None, the current description is preserved. url: If None, the current URL is preserved. events: If None, the current events are preserved. secret... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py
index 8a34c23cd..cbb34f47f 100644
--- a/mlflow/tracking/client.py
+++ b/mlflow/tracking/client.py
@@ -6409,6 +6409,8 @@ class MlflowClient:
if status is not None:
status = WebhookStatus(status) if isinstance(status, str) else sta... | diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py
deleted file mode 100644
index 97fd61b00..000000000
--- a/tests/tracking/test_client_webhooks.py
+++ /dev/null
@@ -1,219 +0,0 @@
-from pathlib import Path
-from typing import Iterator
-
-import pytest
-from cryptography.fernet ... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S8",
"shared_functions": 2,
"shared_lines": 9
},
{
"spec_id": "S12",
"shared_functions": 2,
"shared_lines": 7
},
{
"spec_id": "S11",
"shared_functions": 1,
"shared_lines": 2
}
]
} |
3 | mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1 | {
"spec_id": "S12",
"text": "Raises: mlflow.MlflowException: If the webhook with the specified ID does not exist, if the provided parameters are invalid, or if there are permission issues accessing the webhook.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/tracking/client.py",
... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py
index 8a34c23cd..e8ca4a509 100644
--- a/mlflow/tracking/client.py
+++ b/mlflow/tracking/client.py
@@ -6405,7 +6405,15 @@ class MlflowClient:
A :py:class:`mlflow.entities.webhook.Webhook` object representing the updated webhook.
"""... | diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py
deleted file mode 100644
index 97fd61b00..000000000
--- a/tests/tracking/test_client_webhooks.py
+++ /dev/null
@@ -1,219 +0,0 @@
-from pathlib import Path
-from typing import Iterator
-
-import pytest
-from cryptography.fernet ... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S8",
"shared_functions": 2,
"shared_lines": 7
},
{
"spec_id": "S9",
"shared_functions": 2,
"shared_lines": 7
},
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1 | {
"spec_id": "S13",
"text": "Delete a webhook. This method removes a webhook from the MLflow Model Registry. Once deleted, the webhook will no longer receive notifications for the events it was configured to monitor. Signature: delete_webhook(self, webhook_id: str) -> None. Args: webhook_id: The unique identifier o... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/tracking/client.py b/mlflow/tracking/client.py
index 8a34c23cd..a8b3833e7 100644
--- a/mlflow/tracking/client.py
+++ b/mlflow/tracking/client.py
@@ -6432,6 +6432,7 @@ class MlflowClient:
None
"""
self._get_registry_client().delete_webhook(webhook_id)
+ return we... | diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py
deleted file mode 100644
index 97fd61b00..000000000
--- a/tests/tracking/test_client_webhooks.py
+++ /dev/null
@@ -1,219 +0,0 @@
-from pathlib import Path
-from typing import Iterator
-
-import pytest
-from cryptography.fernet ... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S14",
"shared_functions": 2,
"shared_lines": 2
},
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_client_webhooks.75e39b52.lv1 | {
"spec_id": "S14",
"text": "Raises: MlflowException: If the webhook with the specified ID does not exist, if there are permission issues, or if the deletion operation fails for any other reason.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/tracking/client.py",
"lines": "6437"... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..bd1483000 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,12 +12,12 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_util... | diff --git a/mlflow/tracking/_model_registry/client.py b/mlflow/tracking/_model_registry/client.py
index e1ef46902..def9d64e8 100644
--- a/mlflow/tracking/_model_registry/client.py
+++ b/mlflow/tracking/_model_registry/client.py
@@ -892,6 +892,9 @@ class ModelRegistryClient:
Returns:
None
... | diff --git a/tests/tracking/test_client_webhooks.py b/tests/tracking/test_client_webhooks.py
deleted file mode 100644
index 97fd61b00..000000000
--- a/tests/tracking/test_client_webhooks.py
+++ /dev/null
@@ -1,219 +0,0 @@
-from pathlib import Path
-from typing import Iterator
-
-import pytest
-from cryptography.fernet ... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S13",
"shared_functions": 2,
"shared_lines": 2
},
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S22",
"text": "MlflowException: If the request fails due to timeout, invalid URL, or other exceptions. Also raised if max_retries or backoff_factor exceed their configured limits, or if OAuth authentication is attempted without enabling Databricks SDK.",
"category": "behavioral",
"code_mapping": [... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..d376be9eb 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -248,11 +248,7 @@ def http_request(
**kwargs,
)
except requests.exceptions.Timeout as to:
- raise MlflowException(... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S26",
"shared_functions": 1,
"shared_lines": 12
},
{
"spec_id": "S8",
"shared_functions": 1,
"shared_lines": 12
},
{
"spec_id": "S30",
"shared_functions": 1,
"shared_lines": 5
}
]
} |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S23",
"text": "InvalidUrlException: If the constructed URL is invalid.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/utils/rest_utils.py",
"lines": "254-255",
"function": "http_request"
}
]
} | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..0109d1c44 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -254,7 +254,7 @@ def http_request(
f"{MLFLOW_HTTP_REQUEST_TIMEOUT!s} to a larger value."
) from to
except requests.except... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S26",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S8",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S26",
"text": "Authentication is handled automatically based on the credentials provided in host_creds (username/password, token, client certificates, AWS SigV4, etc.).",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/utils/rest_utils.py",
"lines": "239-268",
"... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..516930bee 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -207,8 +207,6 @@ def http_request(
if host_creds.username and host_creds.password:
basic_auth_str = f"{host_creds.username}:{host_cred... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S8",
"shared_functions": 1,
"shared_lines": 24
},
{
"spec_id": "S22",
"shared_functions": 1,
"shared_lines": 12
},
{
"spec_id": "S30",
"shared_functions": 1,
"shared_lines": 5
},
{
"spec_id... |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S31",
"text": "Validate that total retry timeout is not less than single request timeout. This function performs a validation check to ensure that the total retry timeout duration is not configured to be shorter than the timeout for a single HTTP request. When the total retry timeout is less than the ... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..d270b8bec 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -393,7 +393,7 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second
retry_timeout_seconds: Maximum time for a... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S36",
"shared_functions": 1,
"shared_lines": 9
},
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id": "S35",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id"... |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S32",
"text": "timeout (int | None): Maximum time in seconds to wait for a single HTTP request to complete. If None, no single request timeout validation is performed.",
"category": "structural",
"code_mapping": [
{
"file": "mlflow/utils/rest_utils.py",
"lines": "387",
"funct... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..e066e8dc8 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -392,7 +392,14 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second
timeout: Maximum time for a single HTTP ... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S31",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S33",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S34",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id"... |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S33",
"text": "retry_timeout_seconds (int | None): Maximum time in seconds to spend on all retry attempts combined. If None, no retry timeout validation is performed.",
"category": "structural",
"code_mapping": [
{
"file": "mlflow/utils/rest_utils.py",
"lines": "387",
"functi... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..d8ae255b0 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -392,8 +392,8 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second
timeout: Maximum time for a single HTTP r... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S31",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S32",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S34",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S34",
"text": "None: This function does not return any value. It only performs validation and may issue warnings.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/utils/rest_utils.py",
"lines": "407",
"function": "validate_deployment_timeout_config"
}
]
} | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..8e70ecb9e 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -403,6 +403,7 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second
f"Recommended: Set MLFLOW_DEPLOYM... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S31",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S32",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S33",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S37",
"text": "The function uses stacklevel=2 in the warning to point to the caller's location rather than this validation function",
"category": "implementation_constraint",
"code_mapping": [
{
"file": "mlflow/utils/rest_utils.py",
"lines": "405",
"function": "validate_deplo... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..1cf0dbb1b 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -401,7 +401,7 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second
"premature failures. For long-run... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S31",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S36",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S38",
"shared_functions": 1,
"shared_lines": 1
}
]
} |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S38",
"text": "The warning suggests setting MLFLOW_DEPLOYMENT_PREDICT_TOTAL_TIMEOUT to at least the value of MLFLOW_DEPLOYMENT_PREDICT_TIMEOUT",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/utils/rest_utils.py",
"lines": "399-406",
"function": "validate_deplo... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..b630377d1 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -400,7 +400,7 @@ def validate_deployment_timeout_config(timeout: int | None, retry_timeout_second
"total retry timeout could expire... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S31",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id": "S36",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_id": "S37",
"shared_functions": 1,
"shared_lines": 1
}
]
} |
3 | mlflow__mlflow.93dab383.test_databricks.cbc54e8c.lv1 | {
"spec_id": "S48",
"text": "get_workspace_client creates and caches WorkspaceClient instances with specified configuration including retry and timeout settings",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/utils/rest_utils.py",
"lines": "265-289",
"function": "get_worksp... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
--- a/mlflow/deployments/__init__.py
+++ b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/deployments/__init__.py b/mlflow/deployments/__init__.py
index 9d3545850..33d590fdc 100644
+++ a/mlflow/deployments/__init__.py
--- b/mlflow/deployments/__init__.py
@@ -19,7 +19,7 @@ import json
from mlflow.deployments.base import BaseDeploymentClient
from mlflow.deployments.databricks import Da... | diff --git a/mlflow/utils/rest_utils.py b/mlflow/utils/rest_utils.py
index 368143ad8..64d50e488 100644
--- a/mlflow/utils/rest_utils.py
+++ b/mlflow/utils/rest_utils.py
@@ -259,7 +259,6 @@ def http_request(
raise MlflowException(f"API request to {url} failed with exception {e}")
-@lru_cache(maxsize=5)
def... | diff --git a/tests/deployments/databricks/test_databricks.py b/tests/deployments/databricks/test_databricks.py
deleted file mode 100644
index db1f15091..000000000
--- a/tests/deployments/databricks/test_databricks.py
+++ /dev/null
@@ -1,605 +0,0 @@
-import os
-import warnings
-from unittest import mock
-
-import pytest... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S26",
"shared_functions": 0,
"shared_lines": 4
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S11",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S1",
"text": "Extract score and justification from LLM judge model response text. This function attempts to parse a score (integer) and justification (string) from the response text of an LLM judge model. It supports both JSON format and plain text format with specific patterns. Returns: tuple[int | N... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..b82de8dcc 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -88,7 +88,7 @@ def _extract_score_and_justification(text):
if (match := re.search(r"score: (\d+... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S4",
"shared_functions": 1,
"shared_lines": 27
},
{
"spec_id": "S3",
"shared_functions": 1,
"shared_lines": 9
},
{
"spec_id": "S6",
"shared_functions": 1,
"shared_lines": 7
},
{
"spec_id": ... |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S5",
"text": "The function first normalizes the text by converting \"score\" and \"justification\" keywords to lowercase using case-insensitive regex replacement",
"category": "implementation_constraint",
"code_mapping": [
{
"file": "mlflow/metrics/genai/genai_metric.py",
"lines": ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..771dea1bd 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -76,7 +76,7 @@ def _format_args_string(grading_context_columns: list[str] | None, eval_values,
# Function t... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 1,
"shared_lines": 3
},
{
"spec_id": "S4",
"shared_functions": 1,
"shared_lines": 3
},
{
"spec_id": "S2",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S6",
"text": "If the extracted score is not a number or justification is not a string, returns None for score and an error message for justification. If extraction fails, returns an error message containing the raw output.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/m... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..d2b277890 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -95,6 +95,8 @@ def _extract_score_and_justification(text):
justification = f"Failed to extr... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 1,
"shared_lines": 7
},
{
"spec_id": "S4",
"shared_functions": 1,
"shared_lines": 7
},
{
"spec_id": "S2",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S7",
"text": "def _format_args_string(grading_context_columns: list[str] | None, eval_values, indx) -> str: Format grading context columns into a structured string for LLM evaluation prompts. This function extracts values from specified grading context columns and formats them into a human-readable st... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..23b88abe3 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -67,7 +67,10 @@ def _format_args_string(grading_context_columns: list[str] | None, eval_values,
els... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S8",
"shared_functions": 1,
"shared_lines": 11
},
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 6
},
{
"spec_id": "S9",
"shared_functions": 1,
"shared_lines": 4
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S8",
"text": "Returns an empty string if grading_context_columns is None or empty. Empty or None grading_context_columns will result in an empty string return",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/metrics/genai/genai_metric.py",
"lines": "64-74",
"fu... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..c96cb7d2e 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -63,7 +63,7 @@ def _format_args_string(grading_context_columns: list[str] | None, eval_values,
return... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S7",
"shared_functions": 1,
"shared_lines": 11
},
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S9",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S9",
"text": "Raises: MlflowException: If any column specified in grading_context_columns does not exist in the eval_values dictionary. The exception message will indicate which column is missing and list all available columns.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlf... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..5097476bb 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -56,7 +56,7 @@ def _format_args_string(grading_context_columns: list[str] | None, eval_values,
... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S7",
"shared_functions": 1,
"shared_lines": 4
},
{
"spec_id": "S10",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S8",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S11",
"text": "def make_genai_metric(name: str, definition: str, grading_prompt: str, examples: list[EvaluationExample] | None = None, version: str | None = _get_latest_metric_version(), model: str | None = _get_default_model(), grading_context_columns: str | list[str] | None = None, include_input: bo... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..eb1606e76 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -371,7 +371,7 @@ def make_genai_metric(
include_input: bool = True,
parameters: dict[str, Any] | N... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S13",
"shared_functions": 1,
"shared_lines": 29
},
{
"spec_id": "S15",
"shared_functions": 1,
"shared_lines": 21
},
{
"spec_id": "S14",
"shared_functions": 1,
"shared_lines": 16
},
{
"spec_... |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S12",
"text": "The full grading prompt is stored in the metric_details field of the ``EvaluationMetric`` object. metric_details: (Optional) A description of the metric and how it is calculated.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/metrics/genai/genai_metric.py"... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..4885a4619 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -665,12 +665,16 @@ def make_genai_metric(
# extra params in grading_context_columns can only be passed ... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S11",
"shared_functions": 1,
"shared_lines": 10
},
{
"spec_id": "S19",
"shared_functions": 1,
"shared_lines": 6
},
{
"spec_id": "S31",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id... |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S13",
"text": "Provide example-based few-shot learning for evaluation consistency. Also raised when grading context columns are malformed or when required columns are missing from example grading context.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/metrics/genai/genai... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..b2310a624 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -514,9 +514,12 @@ def make_genai_metric(
elif isinstance(example.grading_context, dict):
... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S11",
"shared_functions": 1,
"shared_lines": 29
},
{
"spec_id": "S18",
"shared_functions": 1,
"shared_lines": 4
},
{
"spec_id": "S12",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id... |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S14",
"text": "Raises: MlflowException: If the specified version is not supported or if there are issues with the evaluation model construction.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/metrics/genai/genai_metric.py",
"lines": "530-545",
"function": "ma... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..07c6f18ee 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -539,7 +539,9 @@ def make_genai_metric(
if examples is not None:
examples = [process_example(e... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S11",
"shared_functions": 1,
"shared_lines": 16
},
{
"spec_id": "S12",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S13",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S15",
"text": "Backward compatibility for metric serialization/deserialization across different MLflow versions. The metric configuration is serialized and stored as an artifact to enable later deserialization and result interpretation. genai_metric_args: (Optional) A dictionary containing arguments s... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..8a776aab8 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -496,7 +496,7 @@ def make_genai_metric(
"aggregations": aggregations,
"greater_is_better":... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S11",
"shared_functions": 1,
"shared_lines": 21
},
{
"spec_id": "S12",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S13",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id... |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S17",
"text": "aggregations: (Optional) The list of options to aggregate the scores. Currently supported options are: min, max, mean, median, variance, p90. Aggregate evaluation results across multiple data points. Generate structured evaluation outputs with per-row scores, justifications, and aggrega... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..35a504924 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -177,7 +177,7 @@ def _get_aggregate_results(scores, aggregations):
"min": np.min,
... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S18",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S21",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S22",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id"... |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S19",
"text": "parameters: (Optional) Parameters for the LLM used to compute the metric. By default, we set the temperature to 0.0, max_tokens to 200, and top_p to 1.0. We recommend setting the temperature to 0.0 for the LLM used as a judge to ensure consistent results.",
"category": "behavioral",
... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/prompts/v1.py b/mlflow/metrics/genai/prompts/v1.py
index 1c86e9814..7f88b16b4 100644
--- a/mlflow/metrics/genai/prompts/v1.py
+++ b/mlflow/metrics/genai/prompts/v1.py
@@ -125,7 +125,7 @@ class EvaluationModel:
grading_prompt=self.grading_prompt,
exampl... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S11",
"shared_functions": 1,
"shared_lines": 7
},
{
"spec_id": "S12",
"shared_functions": 1,
"shared_lines": 6
},
{
"spec_id": "S13",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id"... |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S20",
"text": "def make_genai_metric_from_prompt(name: str, judge_prompt: str | None = None, model: str | None = _get_default_model(), parameters: dict[str, Any] | None = None, aggregations: list[str] | None = None, greater_is_better: bool = True, max_workers: int = 10, metric_metadata: dict[str, Any]... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..41254bba9 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -288,7 +288,15 @@ def make_genai_metric_from_prompt(
"""
import numpy as np
- prompt_template... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S23",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S15",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S21",
"text": "include_input: (Optional) Whether to include the input when computing the metric.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/metrics/genai/genai_metric.py",
"lines": "515-525",
"function": "make_genai_metric.process_example"
},
{
... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..b36efd754 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -526,7 +526,7 @@ def make_genai_metric(
f" Required grading context columns: {grading_conte... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S26",
"shared_functions": 1,
"shared_lines": 3
},
{
"spec_id": "S17",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S18",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S23",
"text": "The prompt may use f-string formatting to include variables. Corresponding variables must be passed as keyword arguments into the resulting metric's eval function. Variables in the judge prompt using f-string formatting must be provided as arguments to the metric's eval function. Raises... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..dc52343cb 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -324,7 +324,7 @@ def make_genai_metric_from_prompt(
kwargs["predictions"] = arg
... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S20",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S15",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id": "S32",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S29",
"text": "def __str__(self): Return a string representation of the EvaluationMetric instance. Returns: str: A formatted string representation of the EvaluationMetric instance in the format \"EvaluationMetric(name=<name>, greater_is_better=<boolean>, ...)\" where additional attributes like long_na... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py
index 75990089f..41c160b66 100644
--- a/mlflow/models/evaluation/base.py
+++ b/mlflow/models/evaluation/base.py
@@ -148,8 +148,7 @@ class EvaluationMetric:
if self.long_name:
parts.append(f"long_name={self.long_name... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S30",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S31",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S30",
"text": "Dynamically generate a GenAIEvaluationMetric class that can be used to evaluate the metric on the given input data. The generated class is callable with a __call__ method that takes the arguments specified in the signature of the eval_fn function. Args: eval_fn: the evaluation function ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/models/evaluation/base.py b/mlflow/models/evaluation/base.py
index 75990089f..984160a9c 100644
--- a/mlflow/models/evaluation/base.py
+++ b/mlflow/models/evaluation/base.py
@@ -275,7 +275,11 @@ def _generate_eval_metric_class(eval_fn, require_strict_signature=False):
) -> MetricValue:
... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S31",
"shared_functions": 1,
"shared_lines": 77
},
{
"spec_id": "S10",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S29",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S32",
"text": "extra_headers: (Optional) Additional headers to be passed to the judge model.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/metrics/genai/genai_metric.py",
"lines": "620-626",
"function": "make_genai_metric.eval_fn"
},
{
"file": ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..d577e0e11 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -333,7 +333,7 @@ def make_genai_metric_from_prompt(
grading_payloads = pd.DataFrame(kwargs).to_dict... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S33",
"shared_functions": 2,
"shared_lines": 10
},
{
"spec_id": "S22",
"shared_functions": 1,
"shared_lines": 7
},
{
"spec_id": "S17",
"shared_functions": 1,
"shared_lines": 0
},
{
"spec_id... |
3 | mlflow__mlflow.93dab383.test_genai_metrics.5f107830.lv1 | {
"spec_id": "S33",
"text": "proxy_url: (Optional) Proxy URL to be used for the judge model. This is useful when the judge model is served via a proxy endpoint, not directly via LLM provider services. If not specified, the default URL for the LLM provider will be used (e.g., https://api.openai.com/v1/chat/completio... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
--- a/mlflow/metrics/__init__.py
+++ b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
-from mlflow.models import (
- ... | diff --git a/mlflow/metrics/__init__.py b/mlflow/metrics/__init__.py
index de8928c86..56c409983 100644
+++ a/mlflow/metrics/__init__.py
--- b/mlflow/metrics/__init__.py
@@ -26,10 +26,10 @@ from mlflow.metrics.metric_definitions import (
_rougeLsum_eval_fn,
_toxicity_eval_fn,
)
+from mlflow.models import (
+ ... | diff --git a/mlflow/metrics/genai/genai_metric.py b/mlflow/metrics/genai/genai_metric.py
index e72e1cb5e..8a6e56321 100644
--- a/mlflow/metrics/genai/genai_metric.py
+++ b/mlflow/metrics/genai/genai_metric.py
@@ -333,7 +333,7 @@ def make_genai_metric_from_prompt(
grading_payloads = pd.DataFrame(kwargs).to_dict... | diff --git a/tests/metrics/genai/test_genai_metrics.py b/tests/metrics/genai/test_genai_metrics.py
deleted file mode 100644
index 5770256f1..000000000
--- a/tests/metrics/genai/test_genai_metrics.py
+++ /dev/null
@@ -1,1387 +0,0 @@
-import inspect
-import re
-from unittest import mock
-
-import numpy as np
-import pand... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S32",
"shared_functions": 2,
"shared_lines": 10
},
{
"spec_id": "S22",
"shared_functions": 1,
"shared_lines": 7
},
{
"spec_id": "S17",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_presigned_url_artifact_repo.8d57288e.lv1 | {
"spec_id": "S17",
"text": "Returns:\n str: A JSON string representation of the Protocol Buffer message with proper formatting (indented with 2 spaces).",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/utils/proto_json_utils.py",
"lines": "127",
"function": "message_to_j... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..e0f99f4b4 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_utils ... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..e0f99f4b4 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_utils ... | diff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py
index 65fb652a7..8dc288245 100644
--- a/mlflow/utils/proto_json_utils.py
+++ b/mlflow/utils/proto_json_utils.py
@@ -126,7 +126,7 @@ def message_to_json(message):
json_dict_with_int64_as_numbers = _merge_json_dicts(
json_dict_... | diff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py
deleted file mode 100644
index 5d5de9527..000000000
--- a/tests/store/artifact/test_presigned_url_artifact_repo.py
+++ /dev/null
@@ -1,363 +0,0 @@
-import json
-import os
-import random
-impo... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S16",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S18",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S19",
"shared_functions": 1,
"shared_lines": 1
}
]
} |
3 | mlflow__mlflow.93dab383.test_presigned_url_artifact_repo.8d57288e.lv1 | {
"spec_id": "S19",
"text": "Uses Google's MessageToJson API internally but applies additional processing to handle int64 fields correctly.\n\nBy default, Google's MessageToJson converts int64 proto fields to JSON strings for compatibility reasons, but this function converts them back to numbers",
"category": "im... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..e0f99f4b4 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_utils ... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..e0f99f4b4 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_utils ... | diff --git a/mlflow/utils/proto_json_utils.py b/mlflow/utils/proto_json_utils.py
index 65fb652a7..e94358bdc 100644
--- a/mlflow/utils/proto_json_utils.py
+++ b/mlflow/utils/proto_json_utils.py
@@ -114,8 +114,9 @@ def message_to_json(message):
# Google's MessageToJson API converts int64 proto fields to JSON strin... | diff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py
deleted file mode 100644
index 5d5de9527..000000000
--- a/tests/store/artifact/test_presigned_url_artifact_repo.py
+++ /dev/null
@@ -1,363 +0,0 @@
-import json
-import os
-import random
-impo... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S16",
"shared_functions": 1,
"shared_lines": 15
},
{
"spec_id": "S30",
"shared_functions": 1,
"shared_lines": 15
},
{
"spec_id": "S18",
"shared_functions": 1,
"shared_lines": 8
},
{
"spec_i... |
3 | mlflow__mlflow.93dab383.test_presigned_url_artifact_repo.8d57288e.lv1 | {
"spec_id": "S21",
"text": "Attempt to execute a function with credential retry logic for handling expired credentials.\n\nThis function implements a retry mechanism for operations that may fail due to credential expiration. It first attempts to execute the provided function with either the original credentials (i... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..e0f99f4b4 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_utils ... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..e0f99f4b4 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_utils ... | diff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py
index bc9f51a24..dd8f4e6a6 100644
--- a/mlflow/store/artifact/artifact_repo.py
+++ b/mlflow/store/artifact/artifact_repo.py
@@ -58,7 +58,7 @@ def _retry_with_new_creds(try_func, creds_func, orig_creds=None):
provided by cr... | diff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py
deleted file mode 100644
index 5d5de9527..000000000
--- a/tests/store/artifact/test_presigned_url_artifact_repo.py
+++ /dev/null
@@ -1,363 +0,0 @@
-import json
-import os
-import random
-impo... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S23",
"shared_functions": 1,
"shared_lines": 6
},
{
"spec_id": "S24",
"shared_functions": 1,
"shared_lines": 6
},
{
"spec_id": "S22",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id"... |
3 | mlflow__mlflow.93dab383.test_presigned_url_artifact_repo.8d57288e.lv1 | {
"spec_id": "S22",
"text": "Returns:\n The return value of try_func when it executes successfully.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/store/artifact/artifact_repo.py",
"lines": "62",
"function": "_retry_with_new_creds"
},
{
"file": "mlflow/sto... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..e0f99f4b4 100644
--- a/mlflow/data/delta_dataset_source.py
+++ b/mlflow/data/delta_dataset_source.py
@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_utils ... | diff --git a/mlflow/data/delta_dataset_source.py b/mlflow/data/delta_dataset_source.py
index a125e5400..e0f99f4b4 100644
+++ a/mlflow/data/delta_dataset_source.py
--- b/mlflow/data/delta_dataset_source.py
@@ -12,7 +12,7 @@ from mlflow.protos.databricks_pb2 import INVALID_PARAMETER_VALUE
from mlflow.utils._spark_utils ... | diff --git a/mlflow/store/artifact/artifact_repo.py b/mlflow/store/artifact/artifact_repo.py
index bc9f51a24..54d16725a 100644
--- a/mlflow/store/artifact/artifact_repo.py
+++ b/mlflow/store/artifact/artifact_repo.py
@@ -59,14 +59,16 @@ def _retry_with_new_creds(try_func, creds_func, orig_creds=None):
"""
try... | diff --git a/tests/store/artifact/test_presigned_url_artifact_repo.py b/tests/store/artifact/test_presigned_url_artifact_repo.py
deleted file mode 100644
index 5d5de9527..000000000
--- a/tests/store/artifact/test_presigned_url_artifact_repo.py
+++ /dev/null
@@ -1,363 +0,0 @@
-import json
-import os
-import random
-impo... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S21",
"shared_functions": 1,
"shared_lines": 2
},
{
"spec_id": "S23",
"shared_functions": 1,
"shared_lines": 1
},
{
"spec_id": "S24",
"shared_functions": 1,
"shared_lines": 1
}
]
} |
3 | mlflow__mlflow.93dab383.test_trace_manager.bb95fbcd.lv1 | {
"spec_id": "S2",
"text": "This method is thread-safe and can be called concurrently from multiple threads. Uses double-checked locking pattern with RLock to minimize synchronization overhead. This method implements the singleton pattern with double-checked locking. _instance_lock = threading.RLock()",
"category... | diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py
index 90d90ab25..4b2aceada 100644
--- a/mlflow/entities/span.py
+++ b/mlflow/entities/span.py
@@ -117,10 +117,10 @@ class Span:
"""The trace ID of the span, a unique identifier for the trace it belongs to."""
return self.get_attribute(Span... | diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py
index 90d90ab25..4b2aceada 100644
+++ a/mlflow/entities/span.py
--- b/mlflow/entities/span.py
@@ -117,10 +117,10 @@ class Span:
"""The trace ID of the span, a unique identifier for the trace it belongs to."""
return self.get_attribute(Span... | diff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py
index 9c9caa149..f75ea28d8 100644
--- a/mlflow/tracing/trace_manager.py
+++ b/mlflow/tracing/trace_manager.py
@@ -59,10 +59,9 @@ class InMemoryTraceManager:
@classmethod
def get_instance(cls):
- if cls._instance is None:
-... | diff --git a/tests/tracing/test_trace_manager.py b/tests/tracing/test_trace_manager.py
deleted file mode 100644
index b5dfcd276..000000000
--- a/tests/tracing/test_trace_manager.py
+++ /dev/null
@@ -1,294 +0,0 @@
-import json
-import time
-from threading import Thread
-
-from mlflow.entities import LiveSpan, Span
-from... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 1,
"shared_lines": 7
},
{
"spec_id": "S17",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_trace_manager.bb95fbcd.lv1 | {
"spec_id": "S17",
"text": "If the trace_id does not exist in the internal traces cache, this will raise a KeyError.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow/tracing/trace_manager.py",
"lines": "115-125",
"function": "InMemoryTraceManager.register_prompt"
}
]
... | diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py
index 90d90ab25..4b2aceada 100644
--- a/mlflow/entities/span.py
+++ b/mlflow/entities/span.py
@@ -117,10 +117,10 @@ class Span:
"""The trace ID of the span, a unique identifier for the trace it belongs to."""
return self.get_attribute(Span... | diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py
index 90d90ab25..4b2aceada 100644
+++ a/mlflow/entities/span.py
--- b/mlflow/entities/span.py
@@ -117,10 +117,10 @@ class Span:
"""The trace ID of the span, a unique identifier for the trace it belongs to."""
return self.get_attribute(Span... | diff --git a/mlflow/tracing/trace_manager.py b/mlflow/tracing/trace_manager.py
index 9c9caa149..d7dfa0f3e 100644
--- a/mlflow/tracing/trace_manager.py
+++ b/mlflow/tracing/trace_manager.py
@@ -111,6 +111,9 @@ class InMemoryTraceManager:
prompt: The prompt version to be registered.
"""
wit... | diff --git a/tests/tracing/test_trace_manager.py b/tests/tracing/test_trace_manager.py
deleted file mode 100644
index b5dfcd276..000000000
--- a/tests/tracing/test_trace_manager.py
+++ /dev/null
@@ -1,294 +0,0 @@
-import json
-import time
-from threading import Thread
-
-from mlflow.entities import LiveSpan, Span
-from... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S1",
"shared_functions": 0,
"shared_lines": 0
},
{
"spec_id": "S2",
"shared_functions": 0,
"shared_lines": 0
}
]
} |
3 | mlflow__mlflow.93dab383.test_trace_manager.bb95fbcd.lv1 | {
"spec_id": "S20",
"text": "When updating the tag, if a current tag value exists, it must be parsed as JSON and validated to be a list type. Invalid JSON or non-list types should raise an MlflowException with a descriptive error message.",
"category": "behavioral",
"code_mapping": [
{
"file": "mlflow... | diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py
index 90d90ab25..4b2aceada 100644
--- a/mlflow/entities/span.py
+++ b/mlflow/entities/span.py
@@ -117,10 +117,10 @@ class Span:
"""The trace ID of the span, a unique identifier for the trace it belongs to."""
return self.get_attribute(Span... | diff --git a/mlflow/entities/span.py b/mlflow/entities/span.py
index 90d90ab25..4b2aceada 100644
+++ a/mlflow/entities/span.py
--- b/mlflow/entities/span.py
@@ -117,10 +117,10 @@ class Span:
"""The trace ID of the span, a unique identifier for the trace it belongs to."""
return self.get_attribute(Span... | diff --git a/mlflow/tracing/utils/prompt.py b/mlflow/tracing/utils/prompt.py
index 8a9847515..3b33fddbd 100644
--- a/mlflow/tracing/utils/prompt.py
+++ b/mlflow/tracing/utils/prompt.py
@@ -23,10 +23,6 @@ def update_linked_prompts_tag(current_tag_value: str | None, prompt_versions: li
if current_tag_value is not No... | diff --git a/tests/tracing/test_trace_manager.py b/tests/tracing/test_trace_manager.py
deleted file mode 100644
index b5dfcd276..000000000
--- a/tests/tracing/test_trace_manager.py
+++ /dev/null
@@ -1,294 +0,0 @@
-import json
-import time
-from threading import Thread
-
-from mlflow.entities import LiveSpan, Span
-from... | {
"case_contract": "Every case is a dict with exactly two keys:\n\n- `args`: a tuple or list of positional arguments for the target callable.\n- `kwargs`: a dict mapping string keys to keyword argument values for the target\n callable.\n\n`check_case(case)` must invoke the target using\n`target(*case[\"args\"], **ca... | {
"preservation_probes": [
{
"spec_id": "S21",
"shared_functions": 1,
"shared_lines": 0
}
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.