Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
id: string
input_sha256: string
utterance_pattern_id: string
selection_rule: struct<kind: string, domain: string, room_id: string, room_predicate: struct<source: string, field: (... 211 chars omitted)
child 0, kind: string
child 1, domain: string
child 2, room_id: string
child 3, room_predicate: struct<source: string, field: string, operator: string, value: string>
child 0, source: string
child 1, field: string
child 2, operator: string
child 3, value: string
child 4, filters: list<item: struct<field: string, operator: string, value: string>>
child 0, item: struct<field: string, operator: string, value: string>
child 0, field: string
child 1, operator: string
child 2, value: string
child 5, rank: struct<field: string, descending: bool, count: int64>
child 0, field: string
child 1, descending: bool
child 2, count: int64
child 6, device_ids: list<item: string>
child 0, item: string
target_device_ids: list<item: string>
child 0, item: string
reference_actions: list<item: struct<device_id: string, service: string, arguments: struct<brightness: int64, level: in (... 253 chars omitted)
child 0, item: struct<device_id: string, service: string, arguments: struct<brightness: int64, level: int64, feedin (... 241 chars omitted)
child 0, device_id: string
child 1, service: string
child 2, arguments: struct<brightness: int64, level: int64, feeding_weight: int64, mode: s
...
child 13, color_mode: string
child 14, target_humidity: int64
expected_tool_trace: list<item: struct<tool_name: string, arguments: string>>
child 0, item: struct<tool_name: string, arguments: string>
child 0, tool_name: string
child 1, arguments: string
implicit_prerequisite_actions: list<item: struct<device_id: string, service: string, arguments: struct<>>>
child 0, item: struct<device_id: string, service: string, arguments: struct<>>
child 0, device_id: string
child 1, service: string
child 2, arguments: struct<>
ground_truth: struct<expected_changes: list<item: struct<device_id: string, path: string, before: string, after: s (... 39 chars omitted)
child 0, expected_changes: list<item: struct<device_id: string, path: string, before: string, after: string>>
child 0, item: struct<device_id: string, path: string, before: string, after: string>
child 0, device_id: string
child 1, path: string
child 2, before: string
child 3, after: string
child 1, preserve_unlisted_state: bool
group_operation: struct<service: string, arguments: struct<brightness: int64, position: int64, speed: string>>
child 0, service: string
child 1, arguments: struct<brightness: int64, position: int64, speed: string>
child 0, brightness: int64
child 1, position: int64
child 2, speed: string
reference_action: null
grounding_reference: null
target_spec_id: string
input_state_sha256: string
to
{'id': Value('string'), 'input_state_sha256': Value('string'), 'ground_truth': {'expected_changes': List({'device_id': Value('string'), 'path': Value('string'), 'before': Json(decode=True), 'after': Json(decode=True)}), 'preserve_unlisted_state': Value('bool'), 'expected_successful_control_calls': Value('int64')}, 'reference_action': {'device_id': Value('string'), 'service': Value('string'), 'arguments': Json(decode=True)}, 'expected_tool_trace': List({'tool_name': Value('string'), 'arguments': Json(decode=True)}), 'target_spec_id': Value('string'), 'grounding_reference': Json(decode=True)}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
id: string
input_sha256: string
utterance_pattern_id: string
selection_rule: struct<kind: string, domain: string, room_id: string, room_predicate: struct<source: string, field: (... 211 chars omitted)
child 0, kind: string
child 1, domain: string
child 2, room_id: string
child 3, room_predicate: struct<source: string, field: string, operator: string, value: string>
child 0, source: string
child 1, field: string
child 2, operator: string
child 3, value: string
child 4, filters: list<item: struct<field: string, operator: string, value: string>>
child 0, item: struct<field: string, operator: string, value: string>
child 0, field: string
child 1, operator: string
child 2, value: string
child 5, rank: struct<field: string, descending: bool, count: int64>
child 0, field: string
child 1, descending: bool
child 2, count: int64
child 6, device_ids: list<item: string>
child 0, item: string
target_device_ids: list<item: string>
child 0, item: string
reference_actions: list<item: struct<device_id: string, service: string, arguments: struct<brightness: int64, level: in (... 253 chars omitted)
child 0, item: struct<device_id: string, service: string, arguments: struct<brightness: int64, level: int64, feedin (... 241 chars omitted)
child 0, device_id: string
child 1, service: string
child 2, arguments: struct<brightness: int64, level: int64, feeding_weight: int64, mode: s
...
child 13, color_mode: string
child 14, target_humidity: int64
expected_tool_trace: list<item: struct<tool_name: string, arguments: string>>
child 0, item: struct<tool_name: string, arguments: string>
child 0, tool_name: string
child 1, arguments: string
implicit_prerequisite_actions: list<item: struct<device_id: string, service: string, arguments: struct<>>>
child 0, item: struct<device_id: string, service: string, arguments: struct<>>
child 0, device_id: string
child 1, service: string
child 2, arguments: struct<>
ground_truth: struct<expected_changes: list<item: struct<device_id: string, path: string, before: string, after: s (... 39 chars omitted)
child 0, expected_changes: list<item: struct<device_id: string, path: string, before: string, after: string>>
child 0, item: struct<device_id: string, path: string, before: string, after: string>
child 0, device_id: string
child 1, path: string
child 2, before: string
child 3, after: string
child 1, preserve_unlisted_state: bool
group_operation: struct<service: string, arguments: struct<brightness: int64, position: int64, speed: string>>
child 0, service: string
child 1, arguments: struct<brightness: int64, position: int64, speed: string>
child 0, brightness: int64
child 1, position: int64
child 2, speed: string
reference_action: null
grounding_reference: null
target_spec_id: string
input_state_sha256: string
to
{'id': Value('string'), 'input_state_sha256': Value('string'), 'ground_truth': {'expected_changes': List({'device_id': Value('string'), 'path': Value('string'), 'before': Json(decode=True), 'after': Json(decode=True)}), 'preserve_unlisted_state': Value('bool'), 'expected_successful_control_calls': Value('int64')}, 'reference_action': {'device_id': Value('string'), 'service': Value('string'), 'arguments': Json(decode=True)}, 'expected_tool_trace': List({'tool_name': Value('string'), 'arguments': Json(decode=True)}), 'target_spec_id': Value('string'), 'grounding_reference': Json(decode=True)}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
EdgeBench-Home
EdgeBench-Home is a frozen bilingual smart-home agent benchmark with aligned Chinese and English task suites.
Dataset size
- 7 task collections (TC1-TC7)
- 60 aligned cases per task collection and language
- 420 Chinese canonical cases
- 420 English canonical cases
- 840 bilingual canonical cases in total
- Every canonical case has both a DR and an EIA evaluation view (1,680 derived view records)
The paper-level dataset size is 840 canonical bilingual cases. DR and EIA are two evaluation interfaces over those cases and must not be counted as additional independent tasks.
Data-only release
This Hugging Face package contains frozen data, interface protocols, manifests, and validation metadata only. It intentionally does not include dataset generators, model runners, environment runtime code, scorers, Judge code, unit tests, or historical review workspaces. The executable evaluation implementation is maintained separately in the EdgeHome evaluation repository.
The deterministic environment and scorer are language-neutral: Chinese and English suites share the same device IDs, service names, state fields, transition semantics, and scoring rules. Natural language prompts and user utterances remain language-specific inside the frozen JSONL files.
Layout
EdgeBench-Home/
zh/
LANGUAGE.json
DR/data/tc1_dr_zh.jsonl ... tc7_dr_zh.jsonl
DR/protocols/dr_tc1.json ... dr_tc7.json
EIA/data/tc1_eia_zh.jsonl ... tc7_eia_zh.jsonl
EIA/protocols/eia_tc1.json ... eia_tc7.json
tc1/data/tc1_zh.jsonl, tc1_eval_zh.jsonl, device_specs.json
...
en/
LANGUAGE.json
DR/data/tc1_dr_en.jsonl ... tc7_dr_en.jsonl
DR/protocols/dr_tc1.json ... dr_tc7.json
EIA/data/tc1_eia_en.jsonl ... tc7_eia_en.jsonl
EIA/protocols/eia_tc1.json ... eia_tc7.json
tc1/data/tc1_en.jsonl, tc1_eval_en.jsonl, device_specs.json
...
Each language directory is an independent suite root. Use EdgeBench-Home/zh for Chinese runs and
EdgeBench-Home/en for English runs.
Model-visible content
For DR inference, send only the model_input value from DR/data/tcX_dr_<lang>.jsonl.
For EIA inference, send only the model_input value from EIA/data/tcX_eia_<lang>.jsonl. The
external evaluation runtime loads the corresponding canonical case and device specifications under
tc1-tc7 while executing tool calls.
Never expose tcX_eval_<lang>.jsonl, outer record IDs, manifests, or validation metadata to the
model. Eval files are frozen ground truth used only by the scorer and TC3 clarification-review path.
Reporting
Chinese and English predictions, Judge reviews, and score reports must be stored and reported separately. Do not combine the two languages into one micro score without also reporting each language independently.
See DATASET_MANIFEST.json and VALIDATION_REPORT.json for frozen hashes and validation results.
- Downloads last month
- 37