Datasets:

asingh15
/

tcs-qwen35-9b-value-trajectories

Tasks:

Text Generation

Tags:

process-reward-model

monte-carlo

theorem-proving

License:

Dataset card Files Files and versions

xet

Community

Dataset Viewer

Duplicate

Split (1)

train

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
schemaVersion: int64
runId: string
generatedAt: string
counts: struct<problems: int64, baseRollouts: int64, trajectories: int64, branchStates: int64, gradeObservat (... 36 chars omitted)
  child 0, problems: int64
  child 1, baseRollouts: int64
  child 2, trajectories: int64
  child 3, branchStates: int64
  child 4, gradeObservations: int64
  child 5, displayedPoints: int64
trendCounts: struct<mostly_increasing: int64, flat: int64, mostly_decreasing: int64, mixed: int64>
  child 0, mostly_increasing: int64
  child 1, flat: int64
  child 2, mostly_decreasing: int64
  child 3, mixed: int64
trendLabels: struct<mostly_increasing: string, flat: string, mostly_decreasing: string, mixed: string>
  child 0, mostly_increasing: string
  child 1, flat: string
  child 2, mostly_decreasing: string
  child 3, mixed: string
trendMethod: struct<values: string, epsilon: double, flatMotion: double, flatEndpointDrift: double, directionalDo (... 16 chars omitted)
  child 0, values: string
  child 1, epsilon: double
  child 2, flatMotion: double
  child 3, flatEndpointDrift: double
  child 4, directionalDominance: double
valueDefinition: struct<branch: string, final: string>
  child 0, branch: string
  child 1, final: string
outputScoreDistribution: struct<population: string, count: int64, mean: double, median: double, standardDeviation: double, bi (... 106 chars omitted)
  child 0, population: string
  child 1, count: int64
  child 2, mean: double
  child 3, median: double
  child 4, standardDev
...
_solutions_per_problem: int64
      child 1, name: string
      child 2, pattern: string
      child 3, prefix_boundary_side: string
      child 4, target_cuts_per_solution: int64
  child 2, judge: struct<all_attempts_judged: bool, hashes: struct<config_sha256: string, index_sha256: string, tokeni (... 94 chars omitted)
      child 0, all_attempts_judged: bool
      child 1, hashes: struct<config_sha256: string, index_sha256: string, tokenizer_sha256: string>
          child 0, config_sha256: string
          child 1, index_sha256: string
          child 2, tokenizer_sha256: string
      child 2, reasoning_effort: string
      child 3, repo: string
      child 4, revision: string
      child 5, served: string
  child 3, policyModel: struct<hashes: struct<config_sha256: string, index_sha256: string, tokenizer_sha256: string>, repo:  (... 41 chars omitted)
      child 0, hashes: struct<config_sha256: string, index_sha256: string, tokenizer_sha256: string>
          child 0, config_sha256: string
          child 1, index_sha256: string
          child 2, tokenizer_sha256: string
      child 1, repo: string
      child 2, revision: string
      child 3, served: string
  child 4, run: struct<base_samples: int64, branch_samples: int64, completed_at: timestamp[s], run_id: string, sourc (... 14 chars omitted)
      child 0, base_samples: int64
      child 1, branch_samples: int64
      child 2, completed_at: timestamp[s]
      child 3, run_id: string
      child 4, source_rows: int64
to
{'complete': Value('bool'), 'counts': {'baseRollouts': Value('int64'), 'branchStates': Value('int64'), 'displayedPoints': Value('int64'), 'gradeObservations': Value('int64'), 'problems': Value('int64'), 'trajectories': Value('int64')}, 'createdAt': Value('string'), 'files': {'problemBundles': {'aggregateSha256': Value('string'), 'bytes': Value('int64'), 'count': Value('int64')}, 'values': {'bytes': Value('int64'), 'path': Value('string'), 'rows': Value('int64'), 'sha256': Value('string')}, 'webIndex': {'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}}, 'inputs': {'baseGrades': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}, 'baseResponses': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}, 'branchGrades': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}, 'branches': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}, 'source': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}}, 'kind': Value('string'), 'outputScoreDistribution': {'binCount': Value('int64'), 'bins': List({'count': Value('int64'), 'lower': Value('float64'), 'upper': Value('float64'), 'upperInclusive': Value('bool')}), 'count': Value('int64'), 'mean': Value('float64'), 'median': Value('float64'), 'population': Value('string'), 'standardDeviation': Value('float64')}, 'runId': Value('string'), 'schemaVersion': Value('int64'), 'trendCounts': {'flat': Value('int64'), 'mixed': Value('int64'), 'mostly_decreasing': Value('int64'), 'mostly_increasing': Value('int64')}, 'trendMethod': {'directionalDominance': Value('float64'), 'epsilon': Value('float64'), 'flatEndpointDrift': Value('float64'), 'flatMotion': Value('float64'), 'values': Value('string')}, 'upstream': {'branchGeneration': {'mode': Value('string'), 'samples_per_row': Value('int64')}, 'branchPolicy': {'base_solutions_per_problem': Value('int64'), 'name': Value('string'), 'pattern': Value('string'), 'prefix_boundary_side': Value('string'), 'target_cuts_per_solution': Value('int64')}, 'judge': {'all_attempts_judged': Value('bool'), 'hashes': {'config_sha256': Value('string'), 'index_sha256': Value('string'), 'tokenizer_sha256': Value('string')}, 'reasoning_effort': Value('string'), 'repo': Value('string'), 'revision': Value('string'), 'served': Value('string')}, 'policyModel': {'hashes': {'config_sha256': Value('string'), 'index_sha256': Value('string'), 'tokenizer_sha256': Value('string')}, 'repo': Value('string'), 'revision': Value('string'), 'served': Value('string')}, 'run': {'base_samples': Value('int64'), 'branch_samples': Value('int64'), 'completed_at': Value('timestamp[s]'), 'run_id': Value('string'), 'source_rows': Value('int64')}}, 'valueDefinition': {'branch': Value('string'), 'final': 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
              schemaVersion: int64
              runId: string
              generatedAt: string
              counts: struct<problems: int64, baseRollouts: int64, trajectories: int64, branchStates: int64, gradeObservat (... 36 chars omitted)
                child 0, problems: int64
                child 1, baseRollouts: int64
                child 2, trajectories: int64
                child 3, branchStates: int64
                child 4, gradeObservations: int64
                child 5, displayedPoints: int64
              trendCounts: struct<mostly_increasing: int64, flat: int64, mostly_decreasing: int64, mixed: int64>
                child 0, mostly_increasing: int64
                child 1, flat: int64
                child 2, mostly_decreasing: int64
                child 3, mixed: int64
              trendLabels: struct<mostly_increasing: string, flat: string, mostly_decreasing: string, mixed: string>
                child 0, mostly_increasing: string
                child 1, flat: string
                child 2, mostly_decreasing: string
                child 3, mixed: string
              trendMethod: struct<values: string, epsilon: double, flatMotion: double, flatEndpointDrift: double, directionalDo (... 16 chars omitted)
                child 0, values: string
                child 1, epsilon: double
                child 2, flatMotion: double
                child 3, flatEndpointDrift: double
                child 4, directionalDominance: double
              valueDefinition: struct<branch: string, final: string>
                child 0, branch: string
                child 1, final: string
              outputScoreDistribution: struct<population: string, count: int64, mean: double, median: double, standardDeviation: double, bi (... 106 chars omitted)
                child 0, population: string
                child 1, count: int64
                child 2, mean: double
                child 3, median: double
                child 4, standardDev
              ...
              _solutions_per_problem: int64
                    child 1, name: string
                    child 2, pattern: string
                    child 3, prefix_boundary_side: string
                    child 4, target_cuts_per_solution: int64
                child 2, judge: struct<all_attempts_judged: bool, hashes: struct<config_sha256: string, index_sha256: string, tokeni (... 94 chars omitted)
                    child 0, all_attempts_judged: bool
                    child 1, hashes: struct<config_sha256: string, index_sha256: string, tokenizer_sha256: string>
                        child 0, config_sha256: string
                        child 1, index_sha256: string
                        child 2, tokenizer_sha256: string
                    child 2, reasoning_effort: string
                    child 3, repo: string
                    child 4, revision: string
                    child 5, served: string
                child 3, policyModel: struct<hashes: struct<config_sha256: string, index_sha256: string, tokenizer_sha256: string>, repo:  (... 41 chars omitted)
                    child 0, hashes: struct<config_sha256: string, index_sha256: string, tokenizer_sha256: string>
                        child 0, config_sha256: string
                        child 1, index_sha256: string
                        child 2, tokenizer_sha256: string
                    child 1, repo: string
                    child 2, revision: string
                    child 3, served: string
                child 4, run: struct<base_samples: int64, branch_samples: int64, completed_at: timestamp[s], run_id: string, sourc (... 14 chars omitted)
                    child 0, base_samples: int64
                    child 1, branch_samples: int64
                    child 2, completed_at: timestamp[s]
                    child 3, run_id: string
                    child 4, source_rows: int64
              to
              {'complete': Value('bool'), 'counts': {'baseRollouts': Value('int64'), 'branchStates': Value('int64'), 'displayedPoints': Value('int64'), 'gradeObservations': Value('int64'), 'problems': Value('int64'), 'trajectories': Value('int64')}, 'createdAt': Value('string'), 'files': {'problemBundles': {'aggregateSha256': Value('string'), 'bytes': Value('int64'), 'count': Value('int64')}, 'values': {'bytes': Value('int64'), 'path': Value('string'), 'rows': Value('int64'), 'sha256': Value('string')}, 'webIndex': {'bytes': Value('int64'), 'path': Value('string'), 'sha256': Value('string')}}, 'inputs': {'baseGrades': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}, 'baseResponses': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}, 'branchGrades': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}, 'branches': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}, 'source': {'bytes': Value('int64'), 'rows': Value('int64'), 'sha256': Value('string')}}, 'kind': Value('string'), 'outputScoreDistribution': {'binCount': Value('int64'), 'bins': List({'count': Value('int64'), 'lower': Value('float64'), 'upper': Value('float64'), 'upperInclusive': Value('bool')}), 'count': Value('int64'), 'mean': Value('float64'), 'median': Value('float64'), 'population': Value('string'), 'standardDeviation': Value('float64')}, 'runId': Value('string'), 'schemaVersion': Value('int64'), 'trendCounts': {'flat': Value('int64'), 'mixed': Value('int64'), 'mostly_decreasing': Value('int64'), 'mostly_increasing': Value('int64')}, 'trendMethod': {'directionalDominance': Value('float64'), 'epsilon': Value('float64'), 'flatEndpointDrift': Value('float64'), 'flatMotion': Value('float64'), 'values': Value('string')}, 'upstream': {'branchGeneration': {'mode': Value('string'), 'samples_per_row': Value('int64')}, 'branchPolicy': {'base_solutions_per_problem': Value('int64'), 'name': Value('string'), 'pattern': Value('string'), 'prefix_boundary_side': Value('string'), 'target_cuts_per_solution': Value('int64')}, 'judge': {'all_attempts_judged': Value('bool'), 'hashes': {'config_sha256': Value('string'), 'index_sha256': Value('string'), 'tokenizer_sha256': Value('string')}, 'reasoning_effort': Value('string'), 'repo': Value('string'), 'revision': Value('string'), 'served': Value('string')}, 'policyModel': {'hashes': {'config_sha256': Value('string'), 'index_sha256': Value('string'), 'tokenizer_sha256': Value('string')}, 'repo': Value('string'), 'revision': Value('string'), 'served': Value('string')}, 'run': {'base_samples': Value('int64'), 'branch_samples': Value('int64'), 'completed_at': Value('timestamp[s]'), 'run_id': Value('string'), 'source_rows': Value('int64')}}, 'valueDefinition': {'branch': Value('string'), 'final': Value('string')}}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

TCS Qwen3.5-9B Value Trajectories

Compact visualization export for tcs_qwen35_9b_gptoss20b_low_vbl_m32_20260812.

Problems: 1,801
Judged base rollouts: 14,408
Selected base trajectories: 3,602
Sampled whitespace cuts: 28,780
Judged branch continuations: 920,960
Policy: Qwen/Qwen3.5-9B
Judge: openai/gpt-oss-20b with low reasoning

For each sampled cut, value is the pooled mean normalized rubric grade across all sampled branches with the same exact problem and reasoning prefix. Each branch has 32 continuations; shared prefixes, including the empty prefix, can therefore have more observations. The terminal point is the normalized rubric grade of the original base trajectory, not another Monte Carlo estimate. data/values.parquet contains one row per displayed point. Browser bundles under web/problems/ contain the exact grade observations and selected trajectory text.

Trend labels use all displayed values, including the terminal rubric score. Changes no larger than 0.05 are ignored. A path is flat when total meaningful motion is at most 0.10 and endpoint drift is at most 0.05. Otherwise, at least two-thirds of meaningful motion must agree with the endpoint direction for a mostly increasing/decreasing label. Remaining trajectories are mixed.

outputScoreDistribution summarizes the normalized terminal rubric scores for all judged base rollouts. It is intentionally broader than the selected set of trajectories used for branched-value analysis.

Downloads last month: 629

Total file size:

138 MB