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
sig: list<item: double>
child 0, item: double
fam: string
note: string
loops: struct<0: struct<cobalt cobalt:10008: list<item: int64>, cobalt cobalt:10278: list<item: int64>, cob (... 212603 chars omitted)
child 0, 0: struct<cobalt cobalt:10008: list<item: int64>, cobalt cobalt:10278: list<item: int64>, cobalt cobalt (... 19231 chars omitted)
child 0, cobalt cobalt:10008: list<item: int64>
child 0, item: int64
child 1, cobalt cobalt:10278: list<item: int64>
child 0, item: int64
child 2, cobalt cobalt:10889: list<item: int64>
child 0, item: int64
child 3, cobalt cobalt:11425: list<item: int64>
child 0, item: int64
child 4, cobalt cobalt:11793: list<item: int64>
child 0, item: int64
child 5, cobalt cobalt:11948: list<item: int64>
child 0, item: int64
child 6, cobalt cobalt:12240: list<item: int64>
child 0, item: int64
child 7, cobalt cobalt:12930: list<item: int64>
child 0, item: int64
child 8, cobalt cobalt:134: list<item: int64>
child 0, item: int64
child 9, cobalt cobalt:13752: list<item: int64>
child 0, item: int64
child 10, cobalt cobalt:143: list<item: int64>
child 0, item: int64
child 11, cobalt cobalt:14918: list<item: int64>
child 0, item: int64
child 12, cobalt cobalt:15834: list<item: int64>
child 0, item: int64
child 13, cobalt cobalt:16205: list<item:
...
, bytes: int64
child 2, paper_b_mining_legs.json: struct<bytes: int64>
child 0, bytes: int64
child 3, paper_b_n3meta.json: struct<bytes: int64>
child 0, bytes: int64
child 4, paper_b_n3n.json: struct<bytes: int64>
child 0, bytes: int64
child 5, paper_b_n3postrl.json: struct<bytes: int64>
child 0, bytes: int64
child 6, paper_b_n3rsa.json: struct<bytes: int64>
child 0, bytes: int64
child 7, paper_b_n3rsa_trajectory.json: struct<bytes: int64>
child 0, bytes: int64
child 8, paper_b_native_difficulty.json: struct<bytes: int64>
child 0, bytes: int64
child 9, paper_b_native_difficulty_rsa.json: struct<bytes: int64>
child 0, bytes: int64
child 10, paper_b_ncp.json: struct<bytes: int64>
child 0, bytes: int64
child 11, paper_b_postrl.json: struct<bytes: int64>
child 0, bytes: int64
child 12, paper_b_regime.json: struct<bytes: int64>
child 0, bytes: int64
child 13, paper_b_regimerl.json: struct<bytes: int64>
child 0, bytes: int64
child 14, paper_b_rsa.json: struct<bytes: int64>
child 0, bytes: int64
child 15, paper_b_rsa_trajectory.json: struct<bytes: int64>
child 0, bytes: int64
child 16, paper_b_students.json: struct<bytes: int64>
child 0, bytes: int64
child 17, paper_b_teacher.json: struct<bytes: int64>
child 0, bytes: int64
child 18, paper_b_tempsweep.json: struct<bytes: int64>
child 0, bytes: int64
generated_utc: timestamp[s]
statcache_bytes: int64
to
{'generated_utc': Value('timestamp[s]'), 'tables': {'paper_b_hightemp.json': {'bytes': Value('int64')}, 'paper_b_mining.json': {'bytes': Value('int64')}, 'paper_b_mining_legs.json': {'bytes': Value('int64')}, 'paper_b_n3meta.json': {'bytes': Value('int64')}, 'paper_b_n3n.json': {'bytes': Value('int64')}, 'paper_b_n3postrl.json': {'bytes': Value('int64')}, 'paper_b_n3rsa.json': {'bytes': Value('int64')}, 'paper_b_n3rsa_trajectory.json': {'bytes': Value('int64')}, 'paper_b_native_difficulty.json': {'bytes': Value('int64')}, 'paper_b_native_difficulty_rsa.json': {'bytes': Value('int64')}, 'paper_b_ncp.json': {'bytes': Value('int64')}, 'paper_b_postrl.json': {'bytes': Value('int64')}, 'paper_b_regime.json': {'bytes': Value('int64')}, 'paper_b_regimerl.json': {'bytes': Value('int64')}, 'paper_b_rsa.json': {'bytes': Value('int64')}, 'paper_b_rsa_trajectory.json': {'bytes': Value('int64')}, 'paper_b_students.json': {'bytes': Value('int64')}, 'paper_b_teacher.json': {'bytes': Value('int64')}, 'paper_b_tempsweep.json': {'bytes': Value('int64')}}, 'statcache_families': Value('int64'), 'statcache_bytes': Value('int64')}
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 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 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
sig: list<item: double>
child 0, item: double
fam: string
note: string
loops: struct<0: struct<cobalt cobalt:10008: list<item: int64>, cobalt cobalt:10278: list<item: int64>, cob (... 212603 chars omitted)
child 0, 0: struct<cobalt cobalt:10008: list<item: int64>, cobalt cobalt:10278: list<item: int64>, cobalt cobalt (... 19231 chars omitted)
child 0, cobalt cobalt:10008: list<item: int64>
child 0, item: int64
child 1, cobalt cobalt:10278: list<item: int64>
child 0, item: int64
child 2, cobalt cobalt:10889: list<item: int64>
child 0, item: int64
child 3, cobalt cobalt:11425: list<item: int64>
child 0, item: int64
child 4, cobalt cobalt:11793: list<item: int64>
child 0, item: int64
child 5, cobalt cobalt:11948: list<item: int64>
child 0, item: int64
child 6, cobalt cobalt:12240: list<item: int64>
child 0, item: int64
child 7, cobalt cobalt:12930: list<item: int64>
child 0, item: int64
child 8, cobalt cobalt:134: list<item: int64>
child 0, item: int64
child 9, cobalt cobalt:13752: list<item: int64>
child 0, item: int64
child 10, cobalt cobalt:143: list<item: int64>
child 0, item: int64
child 11, cobalt cobalt:14918: list<item: int64>
child 0, item: int64
child 12, cobalt cobalt:15834: list<item: int64>
child 0, item: int64
child 13, cobalt cobalt:16205: list<item:
...
, bytes: int64
child 2, paper_b_mining_legs.json: struct<bytes: int64>
child 0, bytes: int64
child 3, paper_b_n3meta.json: struct<bytes: int64>
child 0, bytes: int64
child 4, paper_b_n3n.json: struct<bytes: int64>
child 0, bytes: int64
child 5, paper_b_n3postrl.json: struct<bytes: int64>
child 0, bytes: int64
child 6, paper_b_n3rsa.json: struct<bytes: int64>
child 0, bytes: int64
child 7, paper_b_n3rsa_trajectory.json: struct<bytes: int64>
child 0, bytes: int64
child 8, paper_b_native_difficulty.json: struct<bytes: int64>
child 0, bytes: int64
child 9, paper_b_native_difficulty_rsa.json: struct<bytes: int64>
child 0, bytes: int64
child 10, paper_b_ncp.json: struct<bytes: int64>
child 0, bytes: int64
child 11, paper_b_postrl.json: struct<bytes: int64>
child 0, bytes: int64
child 12, paper_b_regime.json: struct<bytes: int64>
child 0, bytes: int64
child 13, paper_b_regimerl.json: struct<bytes: int64>
child 0, bytes: int64
child 14, paper_b_rsa.json: struct<bytes: int64>
child 0, bytes: int64
child 15, paper_b_rsa_trajectory.json: struct<bytes: int64>
child 0, bytes: int64
child 16, paper_b_students.json: struct<bytes: int64>
child 0, bytes: int64
child 17, paper_b_teacher.json: struct<bytes: int64>
child 0, bytes: int64
child 18, paper_b_tempsweep.json: struct<bytes: int64>
child 0, bytes: int64
generated_utc: timestamp[s]
statcache_bytes: int64
to
{'generated_utc': Value('timestamp[s]'), 'tables': {'paper_b_hightemp.json': {'bytes': Value('int64')}, 'paper_b_mining.json': {'bytes': Value('int64')}, 'paper_b_mining_legs.json': {'bytes': Value('int64')}, 'paper_b_n3meta.json': {'bytes': Value('int64')}, 'paper_b_n3n.json': {'bytes': Value('int64')}, 'paper_b_n3postrl.json': {'bytes': Value('int64')}, 'paper_b_n3rsa.json': {'bytes': Value('int64')}, 'paper_b_n3rsa_trajectory.json': {'bytes': Value('int64')}, 'paper_b_native_difficulty.json': {'bytes': Value('int64')}, 'paper_b_native_difficulty_rsa.json': {'bytes': Value('int64')}, 'paper_b_ncp.json': {'bytes': Value('int64')}, 'paper_b_postrl.json': {'bytes': Value('int64')}, 'paper_b_regime.json': {'bytes': Value('int64')}, 'paper_b_regimerl.json': {'bytes': Value('int64')}, 'paper_b_rsa.json': {'bytes': Value('int64')}, 'paper_b_rsa_trajectory.json': {'bytes': Value('int64')}, 'paper_b_students.json': {'bytes': Value('int64')}, 'paper_b_teacher.json': {'bytes': Value('int64')}, 'paper_b_tempsweep.json': {'bytes': Value('int64')}}, 'statcache_families': Value('int64'), 'statcache_bytes': 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.
Strategic Diversity — derived tables and summary statistics
Machine-readable numbers behind the paper's tables and figures, plus the cached per-problem statistics they are computed from. Every file here is generated; none is hand-transcribed.
tables/
One JSON per paper table. Each carries a source field naming the generator that writes it and a
metric field stating exactly what the numbers mean, so a cell can always be traced back to a
definition rather than guessed at.
| file | contents |
|---|---|
paper_b_students.json, paper_b_n3n.json |
distilled-student pass@k, Qwen3-4B and Nemotron-3-Nano-4B |
paper_b_postrl.json, paper_b_n3postrl.json |
post-RL pass@k |
paper_b_rsa.json, paper_b_n3rsa.json |
RSA pass@1, first and last aggregation iteration |
paper_b_rsa_trajectory.json, paper_b_n3rsa_trajectory.json |
RSA pass@1 at every iteration (0–10) |
paper_b_native_difficulty.json |
pass@1/8/64 by native benchmark difficulty tier |
paper_b_native_difficulty_rsa.json |
RSA by difficulty tier |
paper_b_hightemp.json |
mining-temperature arms |
paper_b_tempsweep.json |
planner x solver temperature grids + IID temperature strip |
paper_b_mining.json, paper_b_mining_legs.json |
mining coverage vs budget; per-leg diagnostics |
paper_b_ncp.json |
next-chapter-prediction coverage vs sample budget |
paper_b_regime.json, paper_b_regimerl.json |
SFT / RFT / ANTI training-regime comparison |
paper_b_teacher.json, paper_b_n3meta.json |
teacher arms; Nemotron run metadata |
Checkpoint selection. Every number reports the best-validation checkpoint, never the final
epoch. Two arms read what looks like a non-best lane and are nonetheless correct: the base model is
untrained, so no checkpoint selection exists; and one arm trained with load_best_model_at_end, so
its final adapter is its best checkpoint (verified byte-identical to the checkpoint the trainer
recorded as best).
statcache/rsa/
Per-family RSA counts, gzipped JSON:
{"sig": [n_files, total_bytes, newest_mtime],
"loops": {"<loop>": {"<suite>\t<problem_id>": [n_correct, n_attempts]}}}
Per-problem granularity is the point: any aggregation — full set, frontier subset, difficulty tier, per-suite, per-iteration — is a groupby over this, so results can be re-derived without the raw shards. Twin problems are included; the paper excludes them, and consumers apply that filter themselves rather than inheriting one analysis's choices.
RSA judges all eleven aggregation iterations at full width; the paper's table reports only the first and last, so the trajectory files and this cache contain strictly more than the table shows.
- Downloads last month
- 35