The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
role: string
content: string
images: null
injected: bool
reasoning_content: string
reasoning_state: null
reasoning_signature: null
reasoning_message_id: string
tool_calls: list<item: struct<id: string, index: int64, function: struct<name: string, arguments: string>, type: (... 9 chars omitted)
child 0, item: struct<id: string, index: int64, function: struct<name: string, arguments: string>, type: string>
child 0, id: string
child 1, index: int64
child 2, function: struct<name: string, arguments: string>
child 0, name: string
child 1, arguments: string
child 3, type: string
name: string
tool_call_id: string
message_id: string
user_display_content: null
token_budget: int64
fate_statement_compat_edits: int64
provider_retry_reasons: string
vibe_exit_code: int64
timed_out: bool
attempt: int64
verdict: string
compile_ok: bool
reason: string
legacy_bigop_replacements: int64
provider_runs: int64
max_rec_depth_added: bool
max_heartbeats_added: bool
complex_abs_replacements: int64
to
{'attempt': Value('int64'), 'token_budget': Value('int64'), 'vibe_exit_code': Value('int64'), 'timed_out': Value('bool'), 'compile_ok': Value('bool'), 'verdict': Value('string'), 'reason': Value('string'), 'legacy_bigop_replacements': Value('int64'), 'complex_abs_replacements': Value('int64'), 'max_rec_depth_added': Value('bool'), 'max_heartbeats_added': Value('bool'), 'fate_statement_compat_edits': Value('int64'), 'provider_runs': Value('int64'), 'provider_retry_reasons': 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 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
role: string
content: string
images: null
injected: bool
reasoning_content: string
reasoning_state: null
reasoning_signature: null
reasoning_message_id: string
tool_calls: list<item: struct<id: string, index: int64, function: struct<name: string, arguments: string>, type: (... 9 chars omitted)
child 0, item: struct<id: string, index: int64, function: struct<name: string, arguments: string>, type: string>
child 0, id: string
child 1, index: int64
child 2, function: struct<name: string, arguments: string>
child 0, name: string
child 1, arguments: string
child 3, type: string
name: string
tool_call_id: string
message_id: string
user_display_content: null
token_budget: int64
fate_statement_compat_edits: int64
provider_retry_reasons: string
vibe_exit_code: int64
timed_out: bool
attempt: int64
verdict: string
compile_ok: bool
reason: string
legacy_bigop_replacements: int64
provider_runs: int64
max_rec_depth_added: bool
max_heartbeats_added: bool
complex_abs_replacements: int64
to
{'attempt': Value('int64'), 'token_budget': Value('int64'), 'vibe_exit_code': Value('int64'), 'timed_out': Value('bool'), 'compile_ok': Value('bool'), 'verdict': Value('string'), 'reason': Value('string'), 'legacy_bigop_replacements': Value('int64'), 'complex_abs_replacements': Value('int64'), 'max_rec_depth_added': Value('bool'), 'max_heartbeats_added': Value('bool'), 'fate_statement_compat_edits': Value('int64'), 'provider_runs': Value('int64'), 'provider_retry_reasons': 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.
Leanstral-1.5 results
This dataset contains preserved evaluation artifacts for
labs-leanstral-1-5 on miniF2F, ProofNet, FATE-M, FATE-H, and PutnamBench. Each trajectory is an interactive
Lean agent run with compiler feedback. The archive includes generated Lean files,
trajectory streams, per-attempt metadata, compilation and SafeVerify logs, and
problem-level summaries.
Results
| Benchmark | Fixed-budget Pass@1 | Fixed-budget Pass@4 | Fixed-budget Pass@8 | Final verified result after extended compute |
|---|---|---|---|---|
| miniF2F | 202/244 (82.8%) | 232/244 (95.1%) | - | 233/244 (95.5%) |
| ProofNet | 39/186 (21.0%) | 42/186 (22.6%) | 99/186 (53.2%) | 113/186 (60.8%) |
| FATE-M | 138/150 (92.0%) | - | - | 147/150 (98.0%) |
| FATE-H | 18/100 (18.0%) | - | - | 28/100 (28.0%) |
| PutnamBench | 58/672 (8.6%) | - | - | 114/672 (17.0%) |
For ProofNet, trajectories 1--4 and 5--8 use the same 4M-token budget; the second group was run only for problems still unsolved, which leaves the cumulative Pass@8 metric unchanged. Two targeted-rerun rows had infrastructure failures, so 99/186 is a strict lower bound. The final 113/186 additionally uses targeted 16M/32M-token rescue trajectories and is not a conventional Pass@k value.
For miniF2F, the fixed-budget setting uses four 2M-token trajectories. The final 233/244 result adds targeted higher-budget rescue on the unresolved remainder and is likewise reported separately from Pass@k.
FATE-M and FATE-H use a 2M-token Pass@1 baseline. Targeted runs allocate four 4M-token trajectories only to baseline failures. FATE-M adds a second group of four 4M-token trajectories to its four remaining failures. The final FATE values are progressive-compute unions, not uniform Pass@4 or Pass@8 estimates. FATE-H contains one released target that does not compile; the table keeps the official 100-item denominator.
PutnamBench reports a complete 672-item Pass@1 baseline with a 2M-token budget. One accepted proof required an independent heartbeat-unlimited revalidation after the statement-locking verifier dropped the candidate's heartbeat option. The archive includes that certificate. A completed targeted rescue then allocated four 4M-token trajectories only to the remaining 614 problems, recovering 56 additional strict certificates. The final 114/672 is therefore a progressive-compute union, not a uniform Pass@4 estimate.
Layout
minif2f/consolidated/: canonical 244-row summary and provenance manifest.minif2f/raw/: all preserved primary, rescue, and compatibility-revalidation runs.proofnet/consolidated/: canonical 186-row summary, provenance manifest, and resource-stage counts.proofnet/raw/: the full baseline, targeted rerun, and targeted rescue artifacts.fate_m/consolidated/: canonical 150-row FATE-M summary and provenance manifest.fate_h/consolidated/: canonical 100-row FATE-H summary and provenance manifest.fate/raw/: FATE-M and FATE-H baseline and targeted-run artifacts.putnam/consolidated/: canonical 672-row PutnamBench summary and provenance manifest.putnam/raw/: the complete Pass@1, heartbeat revalidation, and targeted-rescue artifacts.
The portable archive excludes local symlink views, interrupted runs, preflight
probes, and regenerable Lean build caches. REDACTIONS.md documents one removed
credential value from a saved tool result. No credentials are included.
- Downloads last month
- 95