Safety Edits That Don't Stick: datasets
Collection
ValueShift Research, Lexsi hackathon: RMU unlearning and Antidote repair on Tiny Aya. Repair artifacts are deliberately harmful. • 4 items • Updated
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
id: string
prompt: string
alpha: struct<A3: struct<alpha: null, reason: string>, A4a: struct<alpha: double, harmful_score: double, ss (... 128 chars omitted)
child 0, A3: struct<alpha: null, reason: string>
child 0, alpha: null
child 1, reason: string
child 1, A4a: struct<alpha: double, harmful_score: double, sst2_drop: double>
child 0, alpha: double
child 1, harmful_score: double
child 2, sst2_drop: double
child 2, A5a: struct<alpha: null, reason: string>
child 0, alpha: null
child 1, reason: string
child 3, A6: struct<alpha: double, harmful_score: double, sst2_drop: double>
child 0, alpha: double
child 1, harmful_score: double
child 2, sst2_drop: double
headroom: struct<dose: int64, hs_clean: double, hs_attacked: double, gap: double, go: bool>
child 0, dose: int64
child 1, hs_clean: double
child 2, hs_attacked: double
child 3, gap: double
child 4, go: bool
to
{'headroom': {'dose': Value('int64'), 'hs_clean': Value('float64'), 'hs_attacked': Value('float64'), 'gap': Value('float64'), 'go': Value('bool')}, 'alpha': {'A3': {'alpha': Value('null'), 'reason': Value('string')}, 'A4a': {'alpha': Value('float64'), 'harmful_score': Value('float64'), 'sst2_drop': Value('float64')}, 'A5a': {'alpha': Value('null'), 'reason': Value('string')}, 'A6': {'alpha': Value('float64'), 'harmful_score': Value('float64'), 'sst2_drop': Value('float64')}}}
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
id: string
prompt: string
alpha: struct<A3: struct<alpha: null, reason: string>, A4a: struct<alpha: double, harmful_score: double, ss (... 128 chars omitted)
child 0, A3: struct<alpha: null, reason: string>
child 0, alpha: null
child 1, reason: string
child 1, A4a: struct<alpha: double, harmful_score: double, sst2_drop: double>
child 0, alpha: double
child 1, harmful_score: double
child 2, sst2_drop: double
child 2, A5a: struct<alpha: null, reason: string>
child 0, alpha: null
child 1, reason: string
child 3, A6: struct<alpha: double, harmful_score: double, sst2_drop: double>
child 0, alpha: double
child 1, harmful_score: double
child 2, sst2_drop: double
headroom: struct<dose: int64, hs_clean: double, hs_attacked: double, gap: double, go: bool>
child 0, dose: int64
child 1, hs_clean: double
child 2, hs_attacked: double
child 3, gap: double
child 4, go: bool
to
{'headroom': {'dose': Value('int64'), 'hs_clean': Value('float64'), 'hs_attacked': Value('float64'), 'gap': Value('float64'), 'go': Value('bool')}, 'alpha': {'A3': {'alpha': Value('null'), 'reason': Value('string')}, 'A4a': {'alpha': Value('float64'), 'harmful_score': Value('float64'), 'sst2_drop': Value('float64')}, 'A5a': {'alpha': Value('null'), 'reason': Value('string')}, 'A6': {'alpha': Value('float64'), 'harmful_score': Value('float64'), 'sst2_drop': Value('float64')}}}
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.
Content warning. These are deliberately harmful research artifacts from a safety evaluation. The attack adapters were fine-tuned on harmful BeaverTails answers and make the model comply with harmful requests; the generations include those answers. Use them only to study attacks and defences.
Per-run results for the adapters in ValueShift/exp1-antidote-ru-adapters.
| Folder | What |
|---|---|
results/runs.jsonl |
One summary row per run: StrongREJECT harmful score, coherent-refusal rate, XSTest over-refusal, SST2 and MMLU-Pro accuracy, parse rates, broken flag. timings.jsonl: generation time per run. |
scores/<run_id>.jsonl |
Per-prompt rows with judge score, perplexity, refusal, parsed answer, correctness. |
generations/<run_id>.jsonl |
Greedy completions on StrongREJECT (150), XSTest-safe (250), SST2 (500), MMLU-Pro (200). |
analysis/<config>/ |
summary.md, CSV tables, paired-bootstrap contrasts, plots, decisions.json. |
data_manifest/ |
Ids of the BeaverTails calibration and attack-pool examples. The raw BeaverTails text (and its translations) is not redistributed; the code rebuilds it from the public dataset. |
merged_meta/ |
Repair metadata for merged-weight arms. |