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
text: string
corpora: struct<cc_math: struct<rows: int64, bytes: int64, md5: string, sha256: string>, cc_hq: struct<rows: (... 268 chars omitted)
child 0, cc_math: struct<rows: int64, bytes: int64, md5: string, sha256: string>
child 0, rows: int64
child 1, bytes: int64
child 2, md5: string
child 3, sha256: string
child 1, cc_hq: struct<rows: int64, bytes: int64, md5: string, sha256: string>
child 0, rows: int64
child 1, bytes: int64
child 2, md5: string
child 3, sha256: string
child 2, nemotron_code_github: struct<rows: int64, bytes: int64, md5: string, sha256: string, misses: int64, miss_urls_sample: list (... 15 chars omitted)
child 0, rows: int64
child 1, bytes: int64
child 2, md5: string
child 3, sha256: string
child 4, misses: int64
child 5, miss_urls_sample: list<item: string>
child 0, item: string
child 3, stack_smol_1k: struct<rows: int64, bytes: int64, md5: string, sha256: string>
child 0, rows: int64
child 1, bytes: int64
child 2, md5: string
child 3, sha256: string
revisions: struct<nvidia/Nemotron-Pretraining-Dataset-sample: string, bigcode/the-stack-smol: string>
child 0, nvidia/Nemotron-Pretraining-Dataset-sample: string
child 1, bigcode/the-stack-smol: string
corpora_meta: struct<stack_smol_total_rows: int64, stack_smol_seed: int64, stack_smol_first_idx: list<item: int64> (... 1 chars omitted)
child 0, stack_smol_total_rows: int64
child 1, stack_smol_seed: int64
child 2, stack_smol_first_idx: list<item: int64>
child 0, item: int64
to
{'revisions': {'nvidia/Nemotron-Pretraining-Dataset-sample': Value('string'), 'bigcode/the-stack-smol': Value('string')}, 'corpora': {'cc_math': {'rows': Value('int64'), 'bytes': Value('int64'), 'md5': Value('string'), 'sha256': Value('string')}, 'cc_hq': {'rows': Value('int64'), 'bytes': Value('int64'), 'md5': Value('string'), 'sha256': Value('string')}, 'nemotron_code_github': {'rows': Value('int64'), 'bytes': Value('int64'), 'md5': Value('string'), 'sha256': Value('string'), 'misses': Value('int64'), 'miss_urls_sample': List(Value('string'))}, 'stack_smol_1k': {'rows': Value('int64'), 'bytes': Value('int64'), 'md5': Value('string'), 'sha256': Value('string')}}, 'corpora_meta': {'stack_smol_total_rows': Value('int64'), 'stack_smol_seed': Value('int64'), 'stack_smol_first_idx': List(Value('int64'))}}
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 489, 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 2818, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, 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 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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 2369, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
text: string
corpora: struct<cc_math: struct<rows: int64, bytes: int64, md5: string, sha256: string>, cc_hq: struct<rows: (... 268 chars omitted)
child 0, cc_math: struct<rows: int64, bytes: int64, md5: string, sha256: string>
child 0, rows: int64
child 1, bytes: int64
child 2, md5: string
child 3, sha256: string
child 1, cc_hq: struct<rows: int64, bytes: int64, md5: string, sha256: string>
child 0, rows: int64
child 1, bytes: int64
child 2, md5: string
child 3, sha256: string
child 2, nemotron_code_github: struct<rows: int64, bytes: int64, md5: string, sha256: string, misses: int64, miss_urls_sample: list (... 15 chars omitted)
child 0, rows: int64
child 1, bytes: int64
child 2, md5: string
child 3, sha256: string
child 4, misses: int64
child 5, miss_urls_sample: list<item: string>
child 0, item: string
child 3, stack_smol_1k: struct<rows: int64, bytes: int64, md5: string, sha256: string>
child 0, rows: int64
child 1, bytes: int64
child 2, md5: string
child 3, sha256: string
revisions: struct<nvidia/Nemotron-Pretraining-Dataset-sample: string, bigcode/the-stack-smol: string>
child 0, nvidia/Nemotron-Pretraining-Dataset-sample: string
child 1, bigcode/the-stack-smol: string
corpora_meta: struct<stack_smol_total_rows: int64, stack_smol_seed: int64, stack_smol_first_idx: list<item: int64> (... 1 chars omitted)
child 0, stack_smol_total_rows: int64
child 1, stack_smol_seed: int64
child 2, stack_smol_first_idx: list<item: int64>
child 0, item: int64
to
{'revisions': {'nvidia/Nemotron-Pretraining-Dataset-sample': Value('string'), 'bigcode/the-stack-smol': Value('string')}, 'corpora': {'cc_math': {'rows': Value('int64'), 'bytes': Value('int64'), 'md5': Value('string'), 'sha256': Value('string')}, 'cc_hq': {'rows': Value('int64'), 'bytes': Value('int64'), 'md5': Value('string'), 'sha256': Value('string')}, 'nemotron_code_github': {'rows': Value('int64'), 'bytes': Value('int64'), 'md5': Value('string'), 'sha256': Value('string'), 'misses': Value('int64'), 'miss_urls_sample': List(Value('string'))}, 'stack_smol_1k': {'rows': Value('int64'), 'bytes': Value('int64'), 'md5': Value('string'), 'sha256': Value('string')}}, 'corpora_meta': {'stack_smol_total_rows': Value('int64'), 'stack_smol_seed': Value('int64'), 'stack_smol_first_idx': List(Value('int64'))}}
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.
QATFactory serving-KLD corpora (frozen, md5-pinned)
The frozen neutral corpora behind the standardized serving-KLD measurement
(top-20 forward KL vs a BF16 teacher through real vLLM kernels) used in
qatfactory-experiments
(qwen3.5-27b/RESULTS.md) and specified by the kld-skill in
QATFactory-standard-eval/.cursor/skills/kld-skill/.
These are measurement corpora, not training data. They are frozen: the
files in this repo are the canonical bytes; verify md5 before use. Never
regenerate nemotron_code_github.jsonl — 173/1000 of its pinned GitHub
sources were already dead links at freeze time (2026-07-26).
Files
| file | rows | md5 | provenance |
|---|---|---|---|
cc_math.jsonl |
954 | 0a5e6508ff32256e1d7038f98e865b48 |
nvidia/Nemotron-Pretraining-Dataset-sample : Nemotron-CC-MATH @ 3ad096e6 (full config) |
cc_hq.jsonl |
785 | e35d371c97eb0974f9d17b8bf2335287 |
same dataset : Nemotron-CC-High-Quality @ 3ad096e6 (full config) |
nemotron_code_github.jsonl |
827 | 052142a863e8caa8eec610d77300e29f |
Nemotron-Code-Metadata @ 3ad096e6: 1,000 pinned (repo, commit, path) rows fetched from raw.githubusercontent at the pinned commits; 827 hits / 173 dead |
stack_smol_1k.jsonl |
1000 | 4da554b0e1600c75222dfa4cf43b5482 |
bigcode/the-stack-smol @ 4a6938ce, seed-42 sample of 1,000 files |
MANIFEST.json |
— | — | freeze manifest (revisions, md5s, sha256s, dead-link sample) |
fetch_reasonmix.py |
— | — | reconstructs the fifth corpus (below) for authorized users |
freeze_neutral_corpora.py |
— | — | provenance record of how the four public corpora were frozen (documentation, not reproduction — see the code-corpus caveat) |
The fifth corpus: reasonmix (not redistributed)
reasonmix_qwen36fp8_1k.jsonl (1,000 rows, md5
79fd060960e39605fe502774321d062b) derives from the internal
togethercomputer/Qwen3.5-9B-reasonmix dataset and is therefore not included
here. It is pinned instead: fetch_reasonmix.py rebuilds it byte-exactly
(source file openperfectblend_100k_Qwen3.6-27B-FP8_responses.jsonl @
revision 88de8bf5; indices sorted(random.Random(42).sample(range(n), 1000));
raw lines verbatim; hard md5 assert) for any user whose HF token is authorized
for the togethercomputer org:
HF_TOKEN=<authorized token> python fetch_reasonmix.py
Licenses
The two CC configs come from NVIDIA's Nemotron-Pretraining-Dataset-sample
(CC-BY-4.0). stack_smol_1k files and nemotron_code_github files are
public source code whose licenses remain those of their original repositories
(same posture as bigcode/the-stack-smol); this repo redistributes them solely
as a frozen evaluation corpus with full provenance.
Usage
See kld-skill for the standardized measurement protocol (metric definition
and its approximation, standard kld-core / kld-ladder panels, runner
scripts, replication calibration).
- Downloads last month
- 221