mike-backend / backend /python-service /tests /test_inference_config.py
bardd's picture
Deploy Mike backend-only Space
06671aa verified
Raw History Blame Contribute Delete
4.74 kB
from __future__ import annotations
import json
import os
import sys
import tempfile
import unittest
from pathlib import Path
SERVICE_ROOT = Path(__file__).resolve().parents[1]
BACKEND_ROOT = SERVICE_ROOT.parent
if str(SERVICE_ROOT) not in sys.path:
sys.path.insert(0, str(SERVICE_ROOT))
from inference.config import ( # noqa: E402
InferenceManifestValidationError,
default_main_model,
load_inference_manifest,
reset_inference_manifest_cache,
resolve_manifest_path,
resolve_model,
resolve_timeout_seconds,
)
CANONICAL_MANIFEST = BACKEND_ROOT / "config" / "inference.manifest.json"
class InferenceConfigTests(unittest.TestCase):
def setUp(self) -> None:
reset_inference_manifest_cache()
self._old_manifest_path = os.environ.pop("INFERENCE_MANIFEST_PATH", None)
self._old_model = os.environ.pop("TABULAR_AGENT_MODEL", None)
self._old_timeout = os.environ.pop("TABULAR_AGENT_TIMEOUT_SECONDS", None)
def tearDown(self) -> None:
reset_inference_manifest_cache()
if self._old_manifest_path is not None:
os.environ["INFERENCE_MANIFEST_PATH"] = self._old_manifest_path
else:
os.environ.pop("INFERENCE_MANIFEST_PATH", None)
if self._old_model is not None:
os.environ["TABULAR_AGENT_MODEL"] = self._old_model
else:
os.environ.pop("TABULAR_AGENT_MODEL", None)
if self._old_timeout is not None:
os.environ["TABULAR_AGENT_TIMEOUT_SECONDS"] = self._old_timeout
else:
os.environ.pop("TABULAR_AGENT_TIMEOUT_SECONDS", None)
def test_loads_canonical_manifest(self) -> None:
os.environ["INFERENCE_MANIFEST_PATH"] = str(CANONICAL_MANIFEST)
manifest = load_inference_manifest()
self.assertEqual(manifest["defaults"]["model"], "gemini-3.5-flash")
def test_same_default_model_as_typescript(self) -> None:
os.environ["INFERENCE_MANIFEST_PATH"] = str(CANONICAL_MANIFEST)
self.assertEqual(default_main_model(), "gemini-3.5-flash")
def test_invalid_manifest_fails_validation(self) -> None:
with tempfile.NamedTemporaryFile("w", suffix=".json", delete=False) as handle:
json.dump({"defaults": {"provider": "proxy-openai"}}, handle)
invalid_path = handle.name
try:
with self.assertRaises(InferenceManifestValidationError):
load_inference_manifest(Path(invalid_path))
finally:
os.unlink(invalid_path)
def test_request_model_id_precedence(self) -> None:
os.environ["INFERENCE_MANIFEST_PATH"] = str(CANONICAL_MANIFEST)
os.environ["TABULAR_AGENT_MODEL"] = "gemini-3-flash"
resolved = resolve_model(
request_model_id="gemini-3.5-flash",
feature_id="python-review",
)
self.assertEqual(resolved, "gemini-3.5-flash")
def test_env_model_precedence_over_feature(self) -> None:
os.environ["INFERENCE_MANIFEST_PATH"] = str(CANONICAL_MANIFEST)
os.environ["TABULAR_AGENT_MODEL"] = "gemini-3-flash"
resolved = resolve_model(request_model_id=None, feature_id="python-review")
self.assertEqual(resolved, "gemini-3-flash")
def test_feature_precedence_over_defaults(self) -> None:
os.environ["INFERENCE_MANIFEST_PATH"] = str(CANONICAL_MANIFEST)
resolved = resolve_model(request_model_id=None, feature_id="python-review")
self.assertEqual(resolved, "gemini-3.5-flash")
def test_python_chat_falls_back_to_python_review_model(self) -> None:
os.environ["INFERENCE_MANIFEST_PATH"] = str(CANONICAL_MANIFEST)
resolved = resolve_model(request_model_id=None, feature_id="python-chat")
self.assertEqual(resolved, "gemini-3.5-flash")
def test_legacy_pro_alias_falls_through(self) -> None:
os.environ["INFERENCE_MANIFEST_PATH"] = str(CANONICAL_MANIFEST)
resolved = resolve_model(request_model_id="pro", feature_id="python-review")
self.assertEqual(resolved, "gemini-3.5-flash")
def test_timeout_from_manifest_with_env_override(self) -> None:
os.environ["INFERENCE_MANIFEST_PATH"] = str(CANONICAL_MANIFEST)
self.assertEqual(resolve_timeout_seconds("python-review"), 180.0)
os.environ["TABULAR_AGENT_TIMEOUT_SECONDS"] = "42"
reset_inference_manifest_cache()
self.assertEqual(resolve_timeout_seconds("python-review"), 42.0)
def test_host_development_manifest_fallback_path(self) -> None:
reset_inference_manifest_cache()
os.environ.pop("INFERENCE_MANIFEST_PATH", None)
self.assertEqual(resolve_manifest_path(), CANONICAL_MANIFEST)
if __name__ == "__main__":
unittest.main()