The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
project: string
model: string
evaluation_library: string
auditkit_version: string
generation_library: string
evaluation_mode: string
new_model_calls: int64
new_judge_calls: int64
method_order: list<item: string>
child 0, item: string
runs: struct<safety: struct<baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string (... 441 chars omitted)
child 0, safety: struct<baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string>, static_alpha (... 157 chars omitted)
child 0, baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
child 1, static_alpha4: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
child 2, cast_alpha4: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
child 1, refusal: struct<baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string>, static_alpha (... 157 chars omitted)
child 0, baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate
...
rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
child 2, cast_alpha4: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
source_sha256: struct<heldout_test_manifest.json: string, heldout_test_generations.jsonl: string, heldout_toxic_saf (... 72 chars omitted)
child 0, heldout_test_manifest.json: string
child 1, heldout_test_generations.jsonl: string
child 2, heldout_toxic_safetyguard_v3.jsonl: string
child 3, heldout_hard_refusal_judgments.jsonl: string
limitations: list<item: string>
child 0, item: string
items: list<item: struct<id: string, source: string, category: string, prompt_sha256: string>>
child 0, item: struct<id: string, source: string, category: string, prompt_sha256: string>
child 0, id: string
child 1, source: string
child 2, category: string
child 3, prompt_sha256: string
freeze: struct<model: string, seed: int64, n_per_benchmark: int64, max_new_tokens: int64, config_sha256: str (... 69 chars omitted)
child 0, model: string
child 1, seed: int64
child 2, n_per_benchmark: int64
child 3, max_new_tokens: int64
child 4, config_sha256: string
child 5, condition_vector_sha256: string
child 6, steering_vector_sha256: string
to
{'freeze': {'model': Value('string'), 'seed': Value('int64'), 'n_per_benchmark': Value('int64'), 'max_new_tokens': Value('int64'), 'config_sha256': Value('string'), 'condition_vector_sha256': Value('string'), 'steering_vector_sha256': Value('string')}, 'items': List({'id': Value('string'), 'source': Value('string'), 'category': Value('string'), 'prompt_sha256': Value('string')})}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 544, 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 3052, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 361, 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 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
project: string
model: string
evaluation_library: string
auditkit_version: string
generation_library: string
evaluation_mode: string
new_model_calls: int64
new_judge_calls: int64
method_order: list<item: string>
child 0, item: string
runs: struct<safety: struct<baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string (... 441 chars omitted)
child 0, safety: struct<baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string>, static_alpha (... 157 chars omitted)
child 0, baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
child 1, static_alpha4: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
child 2, cast_alpha4: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
child 1, refusal: struct<baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string>, static_alpha (... 157 chars omitted)
child 0, baseline: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate
...
rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
child 2, cast_alpha4: struct<n: int64, rate: double, count: int64, run_fingerprint: string>
child 0, n: int64
child 1, rate: double
child 2, count: int64
child 3, run_fingerprint: string
source_sha256: struct<heldout_test_manifest.json: string, heldout_test_generations.jsonl: string, heldout_toxic_saf (... 72 chars omitted)
child 0, heldout_test_manifest.json: string
child 1, heldout_test_generations.jsonl: string
child 2, heldout_toxic_safetyguard_v3.jsonl: string
child 3, heldout_hard_refusal_judgments.jsonl: string
limitations: list<item: string>
child 0, item: string
items: list<item: struct<id: string, source: string, category: string, prompt_sha256: string>>
child 0, item: struct<id: string, source: string, category: string, prompt_sha256: string>
child 0, id: string
child 1, source: string
child 2, category: string
child 3, prompt_sha256: string
freeze: struct<model: string, seed: int64, n_per_benchmark: int64, max_new_tokens: int64, config_sha256: str (... 69 chars omitted)
child 0, model: string
child 1, seed: int64
child 2, n_per_benchmark: int64
child 3, max_new_tokens: int64
child 4, config_sha256: string
child 5, condition_vector_sha256: string
child 6, steering_vector_sha256: string
to
{'freeze': {'model': Value('string'), 'seed': Value('int64'), 'n_per_benchmark': Value('int64'), 'max_new_tokens': Value('int64'), 'config_sha256': Value('string'), 'condition_vector_sha256': Value('string'), 'steering_vector_sha256': Value('string')}, 'items': List({'id': Value('string'), 'source': Value('string'), 'category': Value('string'), 'prompt_sha256': Value('string')})}
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.
Tiny Aya Steering: Frozen Evaluation Logs and Scores
Dataset/log repository for the Lexsi Track 2b inference-time safety steering project. Do not assume redistribution rights to OR-Bench's original prompts or full model responses. The intended public release contains aggregate metrics, IDs/categories, judge labels, hash manifests, and sanitized traces only.
Provenance and scope
- Benchmark source:
bench-llms/or-bench,or-bench-toxicandor-bench-hard-1kconfigurations. The original revision was not recorded. - Frozen evaluation: 40 Toxic + 40 Hard prompts, 3 generation methods = 240 generation records.
- Methods: frozen baseline, SafeTune static activation steering, SafeTune CAST.
- Safety label: NVIDIA Nemotron Safety Guard model; refusal label: separate general-purpose NVIDIA Nemotron judge.
- AuditKIT version: 1.1.12; aggregate labels successfully reconstructed without new model calls.
Primary published aggregates
| Method | Toxic judge-flagged unsafe | Hard judge-labeled refusals |
|---|---|---|
| Baseline | 2/40 | 19/40 |
| Static α=4 | 0/40 | 26/40 |
| CAST α=4 | 2/40 | 18/40 |
Interpretation
These are automated judge outputs, not expert ground truth. Hard-set prompts are a proxy for harmless requests. Differences are too small for statistically confident safety claims. Response-aware results include development replay and a fresh routing audit, neither independent safety validation.
Publication safeguards
The owner confirmed redistribution clearance for this release on November 16, 2026. This is an owner attestation, not independent legal verification or a license grant. The linked benchmark card declares CC-BY-4.0; retain applicable attribution and modification notices. No separate license is supplied for this metadata bundle. The label CSV cannot recreate text generation; reproducing counts is distinct from rerunning inference.
A sanitized 160-prompt, 480-generation metadata table and routing summary are included; no raw prompt or response text is distributed. Guard routing judgments are not independently validated safety labels.
- Downloads last month
- 45