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
base_endpoints: struct<count: int64, grade_at_least_half_count: int64, grade_below_half_count: int64, histogram: str (... 179 chars omitted)
child 0, count: int64
child 1, grade_at_least_half_count: int64
child 2, grade_below_half_count: int64
child 3, histogram: struct<counts: list<item: int64>, edges: list<item: double>>
child 0, counts: list<item: int64>
child 0, item: int64
child 1, edges: list<item: double>
child 0, item: double
child 4, max: double
child 5, mean: double
child 6, median: double
child 7, min: double
child 8, p10: double
child 9, p25: double
child 10, p75: double
child 11, p90: double
child 12, std: double
built_at_utc: string
construction: struct<continuation_grades: int64, continuations_per_prefix: int64, prefix_rows: int64, selected_bas (... 119 chars omitted)
child 0, continuation_grades: int64
child 1, continuations_per_prefix: int64
child 2, prefix_rows: int64
child 3, selected_base_trajectories: int64
child 4, semantic_problem_prefixes: int64
child 5, skipped_non_grade_journal_rows: int64
child 6, source_problems: int64
continuation_grades: struct<at_least_half_credit_count: int64, count: int64, full_credit_count: int64, histogram: struct< (... 240 chars omitted)
child 0, at_least_half_credit_count: int64
child 1, count: int64
child 2, full_credit_count: int64
child 3, histogram: struct<counts: list<item: int64>, edges: list<item: double>>
child 0, counts: list<ite
...
14: int64
child 14, 15: int64
child 15, 16: int64
child 16, 17: int64
child 17, 18: int64
child 18, 19: int64
child 19, 20: int64
child 20, 21: int64
child 21, 22: int64
child 22, 23: int64
child 23, 24: int64
child 24, 26: int64
child 25, 27: int64
child 26, 28: int64
child 27, 35: int64
child 28, 38: int64
child 29, 40: int64
child 30, 52: int64
trajectory_patterns: struct<falling: int64, flat: int64, rising: int64, volatile: int64>
child 0, falling: int64
child 1, flat: int64
child 2, rising: int64
child 3, volatile: int64
outputs: struct<browser_problem_count: int64, prefix_parquet_rows: int64, reasoning_trace_count: int64, reaso (... 25 chars omitted)
child 0, browser_problem_count: int64
child 1, prefix_parquet_rows: int64
child 2, reasoning_trace_count: int64
child 3, reasoning_trace_shards: int64
sources: struct<dataset_provenance: struct<bytes: int64, path: string, sha256: string>, prompt_template: stru (... 119 chars omitted)
child 0, dataset_provenance: struct<bytes: int64, path: string, sha256: string>
child 0, bytes: int64
child 1, path: string
child 2, sha256: string
child 1, prompt_template: struct<bytes: int64, path: string, sha256: string>
child 0, bytes: int64
child 1, path: string
child 2, sha256: string
child 2, validation_parquet: struct<bytes: int64, path: string, sha256: string>
child 0, bytes: int64
child 1, path: string
child 2, sha256: string
to
{'built_at_utc': Value('string'), 'outputs': {'browser_problem_count': Value('int64'), 'prefix_parquet_rows': Value('int64'), 'reasoning_trace_count': Value('int64'), 'reasoning_trace_shards': Value('int64')}, 'run_id': Value('string'), 'schema_version': Value('string'), 'sources': {'dataset_provenance': {'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}, 'prompt_template': {'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}, 'validation_parquet': {'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}}, 'target_semantics': {'aggregation': Value('string'), 'endpoint': Value('string'), 'endpoint_values': List(Value('float64')), 'partial': Value('string'), 'partial_value_range': List(Value('float64'))}, 'target_type': Value('string')}
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
base_endpoints: struct<count: int64, grade_at_least_half_count: int64, grade_below_half_count: int64, histogram: str (... 179 chars omitted)
child 0, count: int64
child 1, grade_at_least_half_count: int64
child 2, grade_below_half_count: int64
child 3, histogram: struct<counts: list<item: int64>, edges: list<item: double>>
child 0, counts: list<item: int64>
child 0, item: int64
child 1, edges: list<item: double>
child 0, item: double
child 4, max: double
child 5, mean: double
child 6, median: double
child 7, min: double
child 8, p10: double
child 9, p25: double
child 10, p75: double
child 11, p90: double
child 12, std: double
built_at_utc: string
construction: struct<continuation_grades: int64, continuations_per_prefix: int64, prefix_rows: int64, selected_bas (... 119 chars omitted)
child 0, continuation_grades: int64
child 1, continuations_per_prefix: int64
child 2, prefix_rows: int64
child 3, selected_base_trajectories: int64
child 4, semantic_problem_prefixes: int64
child 5, skipped_non_grade_journal_rows: int64
child 6, source_problems: int64
continuation_grades: struct<at_least_half_credit_count: int64, count: int64, full_credit_count: int64, histogram: struct< (... 240 chars omitted)
child 0, at_least_half_credit_count: int64
child 1, count: int64
child 2, full_credit_count: int64
child 3, histogram: struct<counts: list<item: int64>, edges: list<item: double>>
child 0, counts: list<ite
...
14: int64
child 14, 15: int64
child 15, 16: int64
child 16, 17: int64
child 17, 18: int64
child 18, 19: int64
child 19, 20: int64
child 20, 21: int64
child 21, 22: int64
child 22, 23: int64
child 23, 24: int64
child 24, 26: int64
child 25, 27: int64
child 26, 28: int64
child 27, 35: int64
child 28, 38: int64
child 29, 40: int64
child 30, 52: int64
trajectory_patterns: struct<falling: int64, flat: int64, rising: int64, volatile: int64>
child 0, falling: int64
child 1, flat: int64
child 2, rising: int64
child 3, volatile: int64
outputs: struct<browser_problem_count: int64, prefix_parquet_rows: int64, reasoning_trace_count: int64, reaso (... 25 chars omitted)
child 0, browser_problem_count: int64
child 1, prefix_parquet_rows: int64
child 2, reasoning_trace_count: int64
child 3, reasoning_trace_shards: int64
sources: struct<dataset_provenance: struct<bytes: int64, path: string, sha256: string>, prompt_template: stru (... 119 chars omitted)
child 0, dataset_provenance: struct<bytes: int64, path: string, sha256: string>
child 0, bytes: int64
child 1, path: string
child 2, sha256: string
child 1, prompt_template: struct<bytes: int64, path: string, sha256: string>
child 0, bytes: int64
child 1, path: string
child 2, sha256: string
child 2, validation_parquet: struct<bytes: int64, path: string, sha256: string>
child 0, bytes: int64
child 1, path: string
child 2, sha256: string
to
{'built_at_utc': Value('string'), 'outputs': {'browser_problem_count': Value('int64'), 'prefix_parquet_rows': Value('int64'), 'reasoning_trace_count': Value('int64'), 'reasoning_trace_shards': Value('int64')}, 'run_id': Value('string'), 'schema_version': Value('string'), 'sources': {'dataset_provenance': {'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}, 'prompt_template': {'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}, 'validation_parquet': {'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}}, 'target_semantics': {'aggregation': Value('string'), 'endpoint': Value('string'), 'endpoint_values': List(Value('float64')), 'partial': Value('string'), 'partial_value_range': List(Value('float64'))}, 'target_type': 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.
FineProofs and Polaris Rollout Distributions
Per-prefix aggregate statistics for run
fineproofs_all_qwen35_9b_direct2phase_m32_20260730.
Each row represents one collected response prefix and summarizes exactly 32 fresh
Qwen3.5-9B continuations graded by GPT-OSS-120B against the problem rubric. Every
continuation and endpoint reward is reconstructed exactly as clamped
points / max_points; the rounded journal grade is used only for a consistency
audit with absolute tolerance 0.005. mean_grade is the mean of those canonical
normalized rubric credits in [0, 1]; it is not a Bernoulli success rate. This is the
same dense reward contract declared by fineproofs_prm_mc_context_dataset/v4.
Files
data/prefix_values.parquet: 86,156 prefix rows with trajectory metadata, empirical grade statistics, and a 20-bin continuation-grade histogram.data/summary.json: collection-wide distributions, depth summaries, and PRM-build accounting.data/trace_manifest.json: integrity and coverage metadata for the reasoning export.data/traces/*.json.gz: all 10,454 selected base reasoning trajectories, sharded by the first two hexadecimal characters of the problem ID. Each trace is stored once with exact UTF-16 offsets for every collected prefix step.
The 20 histogram bins partition [0, 1] at intervals of 0.05; grade 1.0 is in the
last bin. The much larger sampled continuation texts and judge records are not
duplicated here.
Trajectory endpoints are joined from the separately frozen GPT-OSS-120B base-grade
journal by (problem_id, sample_index); nullable branch metadata is never used as a label.
Polaris validation sidecar
data/polaris/ contains the matching compact export for the current problem-filtered
Polaris v3 no-context validation split:
prefix_values.parquet: all 1,960 retained prefix points from 238 trajectories.summary.jsonandprovenance.json: distribution, source, and integrity metadata.trace_manifest.jsonandtraces/*.json.gz: all 238 complete reasoning traces from 128 validation problems, with exact UTF-16 offsets for each retained prefix.
Polaris prefix values are binary Monte Carlo success rates over 32 fresh continuations; its terminal marker is exact response correctness. These semantics are intentionally kept distinct from the dense FineProofs rubric grades.
- Downloads last month
- 98