SafetyPulse - data
Collection
SafetyPulse (Lexsi hackathon, Track 2b): eval splits, fine-tune data, contrast pairs, probe bundle, monitoring and audit logs. • 5 items • Updated • 3
Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: CastError
Message: Couldn't cast
method: string
param: double
rand_seed: int64
asr: double
over_refusal: double
task: double
objective: double
params_changed: int64
seed: int64
run: string
hp: struct<epochs: int64, lr: double, batch: int64, accum: int64, save_every: int64>
child 0, epochs: int64
child 1, lr: double
child 2, batch: int64
child 3, accum: int64
child 4, save_every: int64
model: string
condition: string
lora: struct<r: int64, lora_alpha: int64, lora_dropout: double, target_modules: list<item: string>>
child 0, r: int64
child 1, lora_alpha: int64
child 2, lora_dropout: double
child 3, target_modules: list<item: string>
child 0, item: string
dense_until: int64
train_minutes: double
dense_every: null
n_examples: int64
base_dtype: string
log_history: list<item: struct<loss: double, grad_norm: double, learning_rate: double, epoch: double, step: int64 (... 131 chars omitted)
child 0, item: struct<loss: double, grad_norm: double, learning_rate: double, epoch: double, step: int64, train_run (... 119 chars omitted)
child 0, loss: double
child 1, grad_norm: double
child 2, learning_rate: double
child 3, epoch: double
child 4, step: int64
child 5, train_runtime: double
child 6, train_samples_per_second: double
child 7, train_steps_per_second: double
child 8, total_flos: double
child 9, train_loss: double
provenance: struct<git_sha: null, seed: int64, created_at: timestamp[s], torch: string, transformers: string, cu (... 137 chars omitted)
child 0, git_sha: null
child 1, seed: int64
child 2, created_at: timestamp[s]
child 3, torch: string
child 4, transformers: string
child 5, cuda_available: bool
child 6, on_kaggle: bool
child 7, smoke: bool
child 8, peft: string
child 9, gpus: list<item: string>
child 0, item: string
child 10, capability: list<item: int64>
child 0, item: int64
child 11, vram_gb: double
to
{'run': Value('string'), 'condition': Value('string'), 'seed': Value('int64'), 'model': Value('string'), 'hp': {'epochs': Value('int64'), 'lr': Value('float64'), 'batch': Value('int64'), 'accum': Value('int64'), 'save_every': Value('int64')}, 'lora': {'r': Value('int64'), 'lora_alpha': Value('int64'), 'lora_dropout': Value('float64'), 'target_modules': List(Value('string'))}, 'dense_until': Value('int64'), 'dense_every': Value('null'), 'base_dtype': Value('string'), 'n_examples': Value('int64'), 'train_minutes': Value('float64'), 'log_history': List({'loss': Value('float64'), 'grad_norm': Value('float64'), 'learning_rate': Value('float64'), 'epoch': Value('float64'), 'step': Value('int64'), 'train_runtime': Value('float64'), 'train_samples_per_second': Value('float64'), 'train_steps_per_second': Value('float64'), 'total_flos': Value('float64'), 'train_loss': Value('float64')}), 'provenance': {'git_sha': Value('null'), 'seed': Value('int64'), 'created_at': Value('timestamp[s]'), 'torch': Value('string'), 'transformers': Value('string'), 'cuda_available': Value('bool'), 'on_kaggle': Value('bool'), 'smoke': Value('bool'), 'peft': Value('string'), 'gpus': List(Value('string')), 'capability': List(Value('int64')), 'vram_gb': Value('float64')}}
because column names don't match
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1835, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 621, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
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
method: string
param: double
rand_seed: int64
asr: double
over_refusal: double
task: double
objective: double
params_changed: int64
seed: int64
run: string
hp: struct<epochs: int64, lr: double, batch: int64, accum: int64, save_every: int64>
child 0, epochs: int64
child 1, lr: double
child 2, batch: int64
child 3, accum: int64
child 4, save_every: int64
model: string
condition: string
lora: struct<r: int64, lora_alpha: int64, lora_dropout: double, target_modules: list<item: string>>
child 0, r: int64
child 1, lora_alpha: int64
child 2, lora_dropout: double
child 3, target_modules: list<item: string>
child 0, item: string
dense_until: int64
train_minutes: double
dense_every: null
n_examples: int64
base_dtype: string
log_history: list<item: struct<loss: double, grad_norm: double, learning_rate: double, epoch: double, step: int64 (... 131 chars omitted)
child 0, item: struct<loss: double, grad_norm: double, learning_rate: double, epoch: double, step: int64, train_run (... 119 chars omitted)
child 0, loss: double
child 1, grad_norm: double
child 2, learning_rate: double
child 3, epoch: double
child 4, step: int64
child 5, train_runtime: double
child 6, train_samples_per_second: double
child 7, train_steps_per_second: double
child 8, total_flos: double
child 9, train_loss: double
provenance: struct<git_sha: null, seed: int64, created_at: timestamp[s], torch: string, transformers: string, cu (... 137 chars omitted)
child 0, git_sha: null
child 1, seed: int64
child 2, created_at: timestamp[s]
child 3, torch: string
child 4, transformers: string
child 5, cuda_available: bool
child 6, on_kaggle: bool
child 7, smoke: bool
child 8, peft: string
child 9, gpus: list<item: string>
child 0, item: string
child 10, capability: list<item: int64>
child 0, item: int64
child 11, vram_gb: double
to
{'run': Value('string'), 'condition': Value('string'), 'seed': Value('int64'), 'model': Value('string'), 'hp': {'epochs': Value('int64'), 'lr': Value('float64'), 'batch': Value('int64'), 'accum': Value('int64'), 'save_every': Value('int64')}, 'lora': {'r': Value('int64'), 'lora_alpha': Value('int64'), 'lora_dropout': Value('float64'), 'target_modules': List(Value('string'))}, 'dense_until': Value('int64'), 'dense_every': Value('null'), 'base_dtype': Value('string'), 'n_examples': Value('int64'), 'train_minutes': Value('float64'), 'log_history': List({'loss': Value('float64'), 'grad_norm': Value('float64'), 'learning_rate': Value('float64'), 'epoch': Value('float64'), 'step': Value('int64'), 'train_runtime': Value('float64'), 'train_samples_per_second': Value('float64'), 'train_steps_per_second': Value('float64'), 'total_flos': Value('float64'), 'train_loss': Value('float64')}), 'provenance': {'git_sha': Value('null'), 'seed': Value('int64'), 'created_at': Value('timestamp[s]'), 'torch': Value('string'), 'transformers': Value('string'), 'cuda_available': Value('bool'), 'on_kaggle': Value('bool'), 'smoke': Value('bool'), 'peft': Value('string'), 'gpus': List(Value('string')), 'capability': List(Value('int64')), 'vram_gb': Value('float64')}}
because column names don't match
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 1379, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 956, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1702, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1887, 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.
run string | condition string | seed int64 | model string | hp dict | lora dict | dense_until int64 | dense_every null | base_dtype string | n_examples int64 | train_minutes float64 | log_history list | provenance dict |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
alpacacur-s0 | alpacacur | 0 | CohereLabs/tiny-aya-global | {
"epochs": 2,
"lr": 0.0002,
"batch": 2,
"accum": 8,
"save_every": 25
} | {
"r": 16,
"lora_alpha": 32,
"lora_dropout": 0.05,
"target_modules": [
"q_proj",
"k_proj",
"v_proj",
"o_proj",
"gate_proj",
"up_proj",
"down_proj"
]
} | 0 | null | torch.float16 | 1,981 | 17.33256 | [
{
"loss": 2.0397354125976563,
"grad_norm": 0.9192736744880676,
"learning_rate": 0.00011428571428571428,
"epoch": 0.04036326942482341,
"step": 5,
"train_runtime": null,
"train_samples_per_second": null,
"train_steps_per_second": null,
"total_flos": null,
"train_loss": null
}... | {
"git_sha": null,
"seed": 0,
"created_at": "2026-11-07T17:00:38",
"torch": "2.15.0+cu132",
"transformers": "5.22.2",
"cuda_available": true,
"on_kaggle": false,
"smoke": false,
"peft": "0.21.2",
"gpus": [
"NVIDIA A100-SXM4-80GB"
],
"capability": [
8,
0
],
"vram_gb": 85.1
} |
SafetyPulse logs and results
| Folder | Contents |
|---|---|
monitor/ |
CIS (4 variants), refusal spot-check and ASR for all 150 checkpoints, matched bundle; thresholds |
monitor_unpaired_v1/ |
First monitoring pass with the unmatched bundle (superseded) |
audits/audits.jsonl |
Full audits (ASR, over-refusal, GSM8K, MMLU, CIs, per-prompt verdicts) of base, fine-tuned and repaired models |
repair/ |
Dev repair results R1–R6 and the chosen strengths |
report/ |
Final tables (B, C, fine-tune effect), paired comparisons, figure |
specificity/, baseline/ |
Circuit specificity (06c), base-model refusal baseline (04) |
hf_jobs/ |
CircuitKIT cross-check (job 1), Llama Guard 1B vs 8B (job 2), CuratorKIT run (job 3) |
In audits.jsonl, every sampled answer the judge flagged as harmful is replaced by [withheld: judged harmful]; prompts,
verdicts and metrics are complete, so every number in the report can be recomputed.
- Downloads last month
- 47