The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
pipeline_config_hash: string
run_timestamp: string
source_files: list<item: null>
child 0, item: null
stage_counts: struct<JSONLReader: struct<output_count: int64, rejected_count: int64>, HinglishTask: struct<input_c (... 109 chars omitted)
child 0, JSONLReader: struct<output_count: int64, rejected_count: int64>
child 0, output_count: int64
child 1, rejected_count: int64
child 1, HinglishTask: struct<input_count: int64, output_count: int64, rejected_count: int64>
child 0, input_count: int64
child 1, output_count: int64
child 2, rejected_count: int64
child 2, PromptsMetaExporter: struct<exported_count: int64>
child 0, exported_count: int64
rejected_breakdown: struct<generation_parse_failed:HinglishTask: int64>
child 0, generation_parse_failed:HinglishTask: int64
dedup_stats: struct<>
minhash_threshold: null
sampler_stats: null
token_stats: struct<total_prompt_tokens: int64, total_completion_tokens: int64, total_tokens: int64>
child 0, total_prompt_tokens: int64
child 1, total_completion_tokens: int64
child 2, total_tokens: int64
wall_clock_seconds: double
tool_versions: struct<curatorkit: string, python: string>
child 0, curatorkit: string
child 1, python: string
diversity_stats: null
diagnostic_stats: null
diagnostic_files: list<item: null>
child 0, item: null
provenance: struct<schema: string, library: string, version: string, git_sha: null, created_at: timestamp[s], ba (... 277 chars omitted)
child 0, schema: strin
...
: string, ref: string, config: null, split: null, provenance: null>>
child 0, item: struct<kind: string, ref: string, config: null, split: null, provenance: null>
child 0, kind: string
child 1, ref: string
child 2, config: null
child 3, split: null
child 4, provenance: null
child 8, params: struct<generator_model: string, judge_model: null, grpo_scoring_model: null, task_config_hash: strin (... 36 chars omitted)
child 0, generator_model: string
child 1, judge_model: null
child 2, grpo_scoring_model: null
child 3, task_config_hash: string
child 4, ratio_bounds: list<item: double>
child 0, item: double
inputs: list<item: struct<kind: string, ref: string, config: null, split: null, provenance: null>>
child 0, item: struct<kind: string, ref: string, config: null, split: null, provenance: null>
child 0, kind: string
child 1, ref: string
child 2, config: null
child 3, split: null
child 4, provenance: null
version: string
method: string
git_sha: null
base_model: null
schema: string
library: string
created_at: timestamp[s]
params: struct<generator_model: string, judge_model: null, grpo_scoring_model: null, task_config_hash: strin (... 36 chars omitted)
child 0, generator_model: string
child 1, judge_model: null
child 2, grpo_scoring_model: null
child 3, task_config_hash: string
child 4, ratio_bounds: list<item: double>
child 0, item: double
to
{'schema': Value('string'), 'library': Value('string'), 'version': Value('string'), 'git_sha': Value('null'), 'created_at': Value('timestamp[s]'), 'base_model': Value('null'), 'method': Value('string'), 'inputs': List({'kind': Value('string'), 'ref': Value('string'), 'config': Value('null'), 'split': Value('null'), 'provenance': Value('null')}), 'params': {'generator_model': Value('string'), 'judge_model': Value('null'), 'grpo_scoring_model': Value('null'), 'task_config_hash': Value('string'), 'ratio_bounds': List(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
pipeline_config_hash: string
run_timestamp: string
source_files: list<item: null>
child 0, item: null
stage_counts: struct<JSONLReader: struct<output_count: int64, rejected_count: int64>, HinglishTask: struct<input_c (... 109 chars omitted)
child 0, JSONLReader: struct<output_count: int64, rejected_count: int64>
child 0, output_count: int64
child 1, rejected_count: int64
child 1, HinglishTask: struct<input_count: int64, output_count: int64, rejected_count: int64>
child 0, input_count: int64
child 1, output_count: int64
child 2, rejected_count: int64
child 2, PromptsMetaExporter: struct<exported_count: int64>
child 0, exported_count: int64
rejected_breakdown: struct<generation_parse_failed:HinglishTask: int64>
child 0, generation_parse_failed:HinglishTask: int64
dedup_stats: struct<>
minhash_threshold: null
sampler_stats: null
token_stats: struct<total_prompt_tokens: int64, total_completion_tokens: int64, total_tokens: int64>
child 0, total_prompt_tokens: int64
child 1, total_completion_tokens: int64
child 2, total_tokens: int64
wall_clock_seconds: double
tool_versions: struct<curatorkit: string, python: string>
child 0, curatorkit: string
child 1, python: string
diversity_stats: null
diagnostic_stats: null
diagnostic_files: list<item: null>
child 0, item: null
provenance: struct<schema: string, library: string, version: string, git_sha: null, created_at: timestamp[s], ba (... 277 chars omitted)
child 0, schema: strin
...
: string, ref: string, config: null, split: null, provenance: null>>
child 0, item: struct<kind: string, ref: string, config: null, split: null, provenance: null>
child 0, kind: string
child 1, ref: string
child 2, config: null
child 3, split: null
child 4, provenance: null
child 8, params: struct<generator_model: string, judge_model: null, grpo_scoring_model: null, task_config_hash: strin (... 36 chars omitted)
child 0, generator_model: string
child 1, judge_model: null
child 2, grpo_scoring_model: null
child 3, task_config_hash: string
child 4, ratio_bounds: list<item: double>
child 0, item: double
inputs: list<item: struct<kind: string, ref: string, config: null, split: null, provenance: null>>
child 0, item: struct<kind: string, ref: string, config: null, split: null, provenance: null>
child 0, kind: string
child 1, ref: string
child 2, config: null
child 3, split: null
child 4, provenance: null
version: string
method: string
git_sha: null
base_model: null
schema: string
library: string
created_at: timestamp[s]
params: struct<generator_model: string, judge_model: null, grpo_scoring_model: null, task_config_hash: strin (... 36 chars omitted)
child 0, generator_model: string
child 1, judge_model: null
child 2, grpo_scoring_model: null
child 3, task_config_hash: string
child 4, ratio_bounds: list<item: double>
child 0, item: double
to
{'schema': Value('string'), 'library': Value('string'), 'version': Value('string'), 'git_sha': Value('null'), 'created_at': Value('timestamp[s]'), 'base_model': Value('null'), 'method': Value('string'), 'inputs': List({'kind': Value('string'), 'ref': Value('string'), 'config': Value('null'), 'split': Value('null'), 'provenance': Value('null')}), 'params': {'generator_model': Value('string'), 'judge_model': Value('null'), 'grpo_scoring_model': Value('null'), 'task_config_hash': Value('string'), 'ratio_bounds': List(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.
Safety Pencil: data
The processed data behind Safety Pencil (Lexsi AI Safety and Alignment Hackathon, Track 2b): how
fine-tuning Tiny Aya Global on a harmless Hindi task changes its safety behaviour in English, Hindi and Hinglish,
and a closed-form curvature repair. See the repo's WRITEUP.md §4 for how each file was built.
Content warning. This repository contains harmful prompts (requests for dangerous, hateful or illegal content) collected for red-teaming. They are prompts only: no model answers to them are included.
Files
| Folder | What it is | Rows |
|---|---|---|
harmful/{en,hi,hinglish}/{factor,dev,test}.jsonl |
Harmful prompts, split 40 / 10 / 50% per language with a fixed seed, stratified by harm category | en 388 / 97 / 485, hi 362 / 91 / 453, Hinglish 510 / 134 / 635 |
xstest/{en,hi,hinglish}/{dev,test}.jsonl |
XSTest safe prompts (harmless but alarming-sounding), for over-refusal | en and hi 50 / 200, Hinglish 48 / 198 |
task_t1/ |
The benign Hindi fine-tuning task: train_4000 (run A), train_1000 (runs C and D), heldout (300 human-written rows, for task NLL) |
4000 / 1000 / 300 |
mmlu/{en,hi}.jsonl |
Global-MMLU subsets, 200 questions each, the capability-retention check | 200 / 200 |
*/_curatorkit/, */manifest.json, */conversion.json |
CuratorKIT provenance, checksums and conversion settings |
The leak check (data_contamination_check.json in tkaunlaky10/safety-pencil-results) found 0 exact and 0
within-set duplicates between test and fit splits.
Sources and licences
| Data | Source | Licence |
|---|---|---|
| Harmful prompts, en and hi | CohereLabs/aya_redteaming (splits english, hindi) |
Apache-2.0 |
| Harmful prompts, Hinglish | Converted from the en and hi prompts within each split by CohereLabs/aya-expanse-32b through CuratorKIT |
Derived from Apache-2.0 data; the converter model is CC-BY-NC 4.0 |
| Over-refusal prompts | XSTest safe prompts (walledai/XSTest); hi and Hinglish translated by Aya Expanse 32B through CuratorKIT |
CC-BY-4.0 |
| Task T1 | CohereLabs/aya_dataset Hindi rows, topped up from CohereLabs/aya_collection |
Apache-2.0 |
| Capability check | CohereLabs/Global-MMLU (en, hi) |
Apache-2.0 |
Use each part under its source's licence. Run D's fine-tune also used 20 BeaverTails rows (via SafeTune, CC-BY-NC 4.0). They are harmful question–answer pairs and are not included here; get BeaverTails from its original source if you need it.
Intended use
Research on multilingual safety evaluation and repair. Not for training models to produce harmful content.
- Downloads last month
- 35