The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
Error code: DatasetGenerationError
Exception: TypeError
Message: Couldn't cast array of type
struct<is_compliant: bool, constraints: list<item: struct<type: string, description: string, value: string, unit: string>>, evidence: list<item: struct<segment_id: string, quote: string>>, is_coftpliant: bool>
to
{'is_compliant': Value('bool'), 'constraints': List({'type': Value('string'), 'description': Value('string'), 'value': Value('string'), 'unit': Value('string')}), 'evidence': List({'segment_id': Value('string'), 'quote': Value('string')})}
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1887, in _prepare_split_single
writer.write_table(table)
File "/usr/local/lib/python3.12/site-packages/datasets/arrow_writer.py", line 675, in write_table
pa_table = table_cast(pa_table, self._schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2272, in table_cast
return cast_table_to_schema(table, schema)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2224, in cast_table_to_schema
cast_array_to_feature(
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 1795, in wrapper
return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2092, in cast_array_to_feature
raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
TypeError: Couldn't cast array of type
struct<is_compliant: bool, constraints: list<item: struct<type: string, description: string, value: string, unit: string>>, evidence: list<item: struct<segment_id: string, quote: string>>, is_coftpliant: bool>
to
{'is_compliant': Value('bool'), 'constraints': List({'type': Value('string'), 'description': Value('string'), 'value': Value('string'), 'unit': Value('string')}), 'evidence': List({'segment_id': Value('string'), 'quote': Value('string')})}
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1342, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 907, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1736, in _prepare_split
for job_id, done, content in self._prepare_split_single(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/site-packages/datasets/builder.py", line 1919, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
context_tier string | documents list | query string | chosen dict | rejected dict | perturbation dict | detail_elements list | token_count int64 | sample_id string |
|---|---|---|---|---|---|---|---|---|
short | [
{
"doc_id": "GB_T_28053_2011",
"source": "gb_standard_reconstructed",
"segments": [
{
"segment_id": "GB_T_28053_2011_seg_0",
"text": "1 范围.1 残余变形率必须不超过37.8%。\n1 范围.2 残余变形率必须不超过74.5%。\n1 范围.3 残余变形率不应不超过49.8%。\n1 范围.4 爆破压力宜不低于公称工作压力的65.9倍。\n2 规范性引用文件.1 气瓶壁厚宜不小于71.7 MPa。\n2 规范性引用文件.2 ... | A fuel cell vehicle is equipped with Type IV cylinders rated at 10.8 bar. The fuel cell stack operates at -11°C and delivers 2.7 kW. Verify compliance with vehicle hydrogen system standards.
Referenced standards: 呼吸器用复合气瓶, 加氢机 通用技术要求, Energy Storage System Safety and Performance Standard — Emergency Procedures.
Specif... | {
"is_compliant": true,
"constraints": [
{
"type": "tau_4",
"description": "Requirement: .9 m, the equipment is recommended be within the range of 64.4 MPa for Typ",
"value": "is recommended",
"unit": null
},
{
"type": "tau_2",
"description": "Requirement: 文件.8 输氢管道材料... | {
"is_compliant": true,
"constraints": [
{
"type": "tau_4",
"description": "Requirement: .9 m, the equipment is recommended be within the range of 64.4 MPa for Typ",
"value": "is recommended",
"unit": null
},
{
"type": "tau_2",
"description": "Requirement: 文件.8 输氢管道材料... | {
"error_type": "tau_1_threshold",
"original_value": "2.3 MPa",
"perturbed_value": "2.6 MPa",
"detail_element_id": "7e75fe1c380a",
"segment_id": "GB_T_40060_2021_seg_1"
} | [
{
"element_id": "1906543a4059",
"type": "tau_1",
"value": "71.7",
"unit": "MPa",
"span": [
121,
129
],
"segment_id": "GB_T_28053_2011_seg_0",
"quote": "压力的65.9倍。\n2 规范性引用文件.1 气瓶壁厚宜不小于71.7 MPa。\n2 规范性引用文件.2 气瓶壁厚不应不小于51.9 MP",
"level": null
},
{
"element_id": "7... | 4,290 | train_00000 |
short | [{"doc_id":"SYNTH_0060","source":"synthetic","segments":[{"segment_id":"SYNTH_0060_seg_0","text":"1.(...TRUNCATED) | "A stationary storage facility stores hydrogen at 15.1 bar in Type IV vessels operating at 98°C. As(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_2","description":"Requirement: all stationary insta(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_2","description":"Requirement: all stationary insta(...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"7639.8 Nm³","perturbed_value":"9102.3 Nm³","deta(...TRUNCATED) | [{"element_id":"43e992dc06ec","type":"tau_1","value":"0.4","unit":"A","span":[58,63],"segment_id":"S(...TRUNCATED) | 5,900 | train_00001 |
short | [{"doc_id":"SYNTH_0060","source":"synthetic","segments":[{"segment_id":"SYNTH_0060_seg_0","text":"1.(...TRUNCATED) | "A fuel cell vehicle is equipped with Type III cylinders rated at 18.7 bar. The fuel cell stack oper(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_2","description":"Requirement: recommended not be (...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_2","description":"Requirement: recommended not be (...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"35 MPa","perturbed_value":"39 MPa","detail_element(...TRUNCATED) | [{"element_id":"43e992dc06ec","type":"tau_1","value":"0.4","unit":"A","span":[58,63],"segment_id":"S(...TRUNCATED) | 8,121 | train_00002 |
short | [{"doc_id":"GB_T_18487_1_2015","source":"gb_standard_reconstructed","segments":[{"segment_id":"GB_T_(...TRUNCATED) | "A hydrogen refuelling station dispenses hydrogen at 63.3 MPa with a flow rate of 186.2 Nm³/h. The (...TRUNCATED) | {"is_compliant":false,"constraints":[{"type":"tau_4","description":"Requirement: competent authority(...TRUNCATED) | {"is_compliant":false,"constraints":[{"type":"tau_4","description":"Requirement: competent authority(...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"2 kW","perturbed_value":"2 kW","detail_element_id"(...TRUNCATED) | [{"element_id":"46bc61685736","type":"tau_1","value":"14.6","unit":"Ω","span":[81,87],"segment_id":(...TRUNCATED) | 448,427 | train_00003 |
medium | [{"doc_id":"SYNTH_0111","source":"synthetic","segments":[{"segment_id":"SYNTH_0111_seg_0","text":"2.(...TRUNCATED) | "An electrolyser plant produces hydrogen at a purity of 99.579% and delivers it at 97.9 MPa to a ref(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_2","description":"Requirement: commended be maintai(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_2","description":"Requirement: commended be maintai(...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"8.2 W","perturbed_value":"7.2 W","detail_element_i(...TRUNCATED) | [{"element_id":"e3c0a86839b7","type":"tau_1","value":"27.0","unit":"mm","span":[133,140],"segment_id(...TRUNCATED) | 19,462 | train_00004 |
short | [{"doc_id":"SYNTH_0008","source":"synthetic","segments":[{"segment_id":"SYNTH_0008_seg_0","text":"1.(...TRUNCATED) | "A fleet operator transports compressed hydrogen via tube trailers at 93.6 bar, with each trailer ca(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_4","description":"Requirement: \nA.8 the equipment(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_4","description":"Requirement: \nA.8 the equipment(...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"3 W","perturbed_value":"4 W","detail_element_id":"(...TRUNCATED) | [{"element_id":"6214025d33c9","type":"tau_1","value":"414.9","unit":"ppm","span":[42,51],"segment_id(...TRUNCATED) | 8,068 | train_00005 |
short | [{"doc_id":"GB_T_33145_2016","source":"gb_standard_reconstructed","segments":[{"segment_id":"GB_T_33(...TRUNCATED) | "A stationary storage facility stores hydrogen at 96.0 MPa in Type II vessels operating at 117°C. A(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_3","description":"Requirement: 2963.4 kW for indoor(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_3","description":"Requirement: 2963.4 kW for indoor(...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"2 W","perturbed_value":"2 W","detail_element_id":"(...TRUNCATED) | [{"element_id":"8d35ee0c4416","type":"tau_1","value":"81.9","unit":"°C","span":[14,20],"segment_id"(...TRUNCATED) | 6,218 | train_00006 |
short | [{"doc_id":"EURLEX_32018L0844","source":"eurlex","segments":[{"segment_id":"EURLEX_32018L0844_seg_0"(...TRUNCATED) | "A rooftop installation facility stores hydrogen at 69.6 bar in composite overwrapped vessels operat(...TRUNCATED) | {"is_compliant":false,"constraints":[{"type":"tau_5","description":"Requirement: systems, if techni(...TRUNCATED) | {"is_compliant":false,"constraints":[{"type":"tau_5","description":"Requirement: systems, if techni(...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"22 m","perturbed_value":"25 m","detail_element_id"(...TRUNCATED) | [{"element_id":"0df38e1be741","type":"tau_1","value":"75","unit":"L","span":[108,112],"segment_id":"(...TRUNCATED) | 77,261 | train_00007 |
medium | [{"doc_id":"SYNTH_0123","source":"synthetic","segments":[{"segment_id":"SYNTH_0123_seg_0","text":"1.(...TRUNCATED) | "A fleet operator transports compressed hydrogen via tube trailers at 62.0 MPa, with each trailer ca(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_4","description":"Requirement: \n8.1 the equipment(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_4","description":"Requirement: \n8.1 the equipment(...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"2 W","perturbed_value":"2 W","detail_element_id":"(...TRUNCATED) | [{"element_id":"eb7795efb306","type":"tau_1","value":"332.7","unit":"Nm³","span":[30,39],"segment_i(...TRUNCATED) | 16,878 | train_00008 |
short | [{"doc_id":"CFR_49_199","source":"us_cfr","segments":[{"segment_id":"CFR_49_199_seg_0","text":"§ 19(...TRUNCATED) | "A hydrogen refuelling station dispenses hydrogen at 75.1 bar with a flow rate of 4.7 kg/h. The stat(...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_2","description":"Requirement: least six tests in (...TRUNCATED) | {"is_compliant":true,"constraints":[{"type":"tau_2","description":"Requirement: least six tests in (...TRUNCATED) | {"error_type":"tau_1_threshold","original_value":"50 %","perturbed_value":"41 %","detail_element_id"(...TRUNCATED) | [{"element_id":"98842ad753ec","type":"tau_1","value":"199.2","unit":"A","span":[242,249],"segment_id(...TRUNCATED) | 12,170 | train_00009 |
DetailBench
A Benchmark for Detail Hallucination in Long Regulatory Documents
DetailBench is a benchmark for evaluating and mitigating detail hallucination in LLM outputs on long regulatory documents.
Overview
Large language models frequently produce detail hallucinations—subtle errors in threshold values, units, scopes, obligation levels, and conditions—when processing long regulatory documents. DetailBench provides:
- 322 source documents (172 real + 150 synthetic) from three jurisdictions
- 13,000 preference pairs (10,000 train / 1,000 validation / 2,000 test)
- Five detail error types (τ₁–τ₅) with balanced training distribution
- Three context-length tiers: Short (8K–16K), Medium (16K–32K), Long (32K–64K tokens)
Data Sources
| Source | Count | Description |
|---|---|---|
| GB Standards | 65 | Chinese national standards on hydrogen production, storage, transportation, and safety |
| US CFR | 31 | Code of Federal Regulations (Title 49: Transportation, Title 40: Environmental Protection) via eCFR API |
| EUR-Lex | 76 | EU regulations on hydrogen infrastructure, clean energy, pressure equipment via CELLAR API |
| Synthetic | 150 | Domain-template generated documents for training augmentation |
Schema
Each sample in the JSONL files contains:
{
"sample_id": "test_00000",
"context_tier": "long",
"token_count": 43368,
"documents": [
{
"doc_id": "SYNTH_0075",
"source": "synthetic",
"segments": [
{
"segment_id": "SYNTH_0075_seg_0",
"text": "...",
"token_count": 605
}
]
}
],
"query": "An electrolyser plant produces hydrogen at ...",
"chosen": {
"is_compliant": true,
"constraints": [
{"type": "tau_1", "description": "...", "value": "82", "unit": "°C"}
],
"evidence": [
{"segment_id": "...", "quote": "..."}
]
},
"rejected": {
"is_compliant": true,
"constraints": ["... (with one perturbed detail)"],
"evidence": ["..."]
},
"perturbation": {
"error_type": "tau_5_condition",
"original_value": "where appropriate",
"perturbed_value": "[dropped]",
"detail_element_id": "...",
"segment_id": "..."
},
"detail_elements": [
{
"element_id": "...",
"type": "tau_1",
"value": "3928.0",
"unit": "kg",
"span": [46, 55],
"segment_id": "...",
"quote": "..."
}
]
}
Detail Error Taxonomy
| Type | Name | Description | Example |
|---|---|---|---|
| τ₁ | Threshold | Numeric value errors | "pressure ≤ 35 MPa" → "pressure ≤ 45 MPa" |
| τ₂ | Unit | Measurement unit errors | "distance in meters" → "distance in feet" |
| τ₃ | Scope | Applicability scope errors | "for indoor facilities" → "for all facilities" |
| τ₄ | Level | Obligation level errors | "shall comply" → "should comply" |
| τ₅ | Condition | Conditional clause errors | "if temperature exceeds 60°C" → condition dropped |
Evaluation Metrics
- Compliance Accuracy: Fraction of correct compliance judgments
- Detail Error Rate (DER): Per-type and overall error rate on detail elements
- Evidence F1: Precision/recall/F1 of predicted evidence citations
- Evidence Consistency: Fraction of citations where quoted text matches source
Usage
from datasets import load_dataset
ds = load_dataset("YOUR_USERNAME/DetailBench")
# Access splits
train = ds["train"] # 10,000 samples
val = ds["validation"] # 1,000 samples
test = ds["test"] # 2,000 samples
# Example: inspect a test sample
sample = test[0]
print(sample["query"])
print(sample["context_tier"]) # "short", "medium", or "long"
print(len(sample["documents"]))
Split Statistics
| Split | Samples | Short | Medium | Long |
|---|---|---|---|---|
| Train | 10,000 | 6,463 | 2,263 | 1,274 |
| Val | 1,000 | 605 | 249 | 146 |
| Test | 2,000 | 1,215 | 519 | 266 |
Error type distribution in the training set is balanced at 20% each (2,000 per type).
License
This dataset is released under CC-BY-4.0. The underlying regulatory documents are sourced from public government repositories (eCFR, EUR-Lex, openstd.samr.gov.cn).
- Downloads last month
- 26