Evals-Runs-Datasets
Collection
Datasets of evaluation runs • 2 items • Updated
Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
prompt_id: string
model: string
judge: string
refused: bool
challenged: bool
endorsement: int64
rationale: string
ts: timestamp[s]
usage: null
finish_reason: string
response: string
model_id: string
latency_s: double
error: string
prompt: string
to
{'model': Value('string'), 'model_id': Value('string'), 'prompt_id': Value('string'), 'prompt': Value('string'), 'response': Value('string'), 'finish_reason': Value('string'), 'usage': Json(decode=True), 'latency_s': Value('float64'), 'ts': Value('timestamp[s]'), 'error': Value('string')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, 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 127, 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 483, 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 2951, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, 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 547, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, 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 343, 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 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
prompt_id: string
model: string
judge: string
refused: bool
challenged: bool
endorsement: int64
rationale: string
ts: timestamp[s]
usage: null
finish_reason: string
response: string
model_id: string
latency_s: double
error: string
prompt: string
to
{'model': Value('string'), 'model_id': Value('string'), 'prompt_id': Value('string'), 'prompt': Value('string'), 'response': Value('string'), 'finish_reason': Value('string'), 'usage': Json(decode=True), 'latency_s': Value('float64'), 'ts': Value('timestamp[s]'), 'error': 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.
Propaganda-Bait-Eval: runs
Raw model outputs from runs of Propaganda-Bait-Eval. The prompts themselves are a separate dataset: danielrosehill/Propaganda-Bait-Eval. This dataset holds only what models said back, plus judge scores where a run was judged.
These are unfiltered model responses to loaded-premise prompts. Some repeat state propaganda. They are published as evidence for evaluation research.
Runs
smoke-01 (2026-10-19)
A harness check: 18 prompts and 2 models, judged. It contains responses.jsonl, judge.jsonl, scores.jsonl and the report (report.md and report.json).
on-air-01 (2026-10-19)
A small first look, not judged. It used 5 "explain in about a minute" prompts, written for this run and not taken from the dataset. They were Tiananmen in English and Chinese, Bucha in English and Russian, and Khashoggi in English, sent to 7 models with no system prompt:
- Through Hugging Face Inference Providers:
deepseek-ai/DeepSeek-V4-Flash,zai-org/GLM-5.3andmoonshotai/Kimi-K2-Instruct-0905. - On dedicated HF Inference Endpoints (vLLM on an L40S, deleted after the run):
yandex/YandexGPT-5-Lite-8B-instruct,ai-sage/GigaChat3.1-10B-A1.8Bandhumain-ai/ALLaM-7B-Instruct-preview. - The control, through OpenRouter:
openai/gpt-6.1-sol.
Files: on-air-01/responses.jsonl. Each row holds model, model_id, prompt_id, prompt, response, finish_reason, usage and latency_s. Errored attempts (HF 429s and 404s) are kept as rows with response: null. Harness: pbe/onair.py. The endpoint recipe is in docs/hf-inference-endpoints.md in the GitHub repo.
Headline: the answers change with the question's language. Kimi K2 gave the Tiananmen death toll in English and a Party statement in Chinese. YandexGPT named Russian forces for Bucha in English, but named no one in Russian.
- Downloads last month
- 30