Hugging Face's logo

Datasets:
vital-ar
/
adversarial-physics-eval-traces-benchmarks

Tasks:
Text Generation
Question Answering
Text Classification
Languages:
English
Tags:
physics
benchmark
evaluation-traces
multi-turn
curatorkit
aligntune
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
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
metadata: struct<validation_source: string, validation_local_path: string, limit: null, batch_size: int64, num (... 266 chars omitted)
  child 0, validation_source: string
  child 1, validation_local_path: string
  child 2, limit: null
  child 3, batch_size: int64
  child 4, num_gpu_workers: int64
  child 5, tier: string
  child 6, curated_first: bool
  child 7, base_model: string
  child 8, solver_adapter: string
  child 9, verifier_adapter: string
  child 10, prompt_mode: string
  child 11, solver_max_new_tokens: int64
  child 12, verifier_max_new_tokens: int64
  child 13, temperature_solver: double
  child 14, temperature_verifier: double
metrics: struct<total_problems: int64, max_iterations: int64, solver: struct<turn0_correct: int64, turn0_accu (... 3067 chars omitted)
  child 0, total_problems: int64
  child 1, max_iterations: int64
  child 2, solver: struct<turn0_correct: int64, turn0_accuracy: double, final_correct: int64, final_accuracy: double, n (... 566 chars omitted)
      child 0, turn0_correct: int64
      child 1, turn0_accuracy: double
      child 2, final_correct: int64
      child 3, final_accuracy: double
      child 4, net_revision_gain: double
      child 5, turn0_sentinel_compliance_rate: double
      child 6, final_sentinel_compliance_rate: double
      child 7, converted_count: int64
      child 8, turn0_wrong_count: int64
      child 9, correction_conversion_rate: double
      child 10, regressed_count: int64
      child 11, regression_rate: double
 
...
4, turn: int64, accumulated_directives: list<item: string>, solver_prompt: str (... 556 chars omitted)
      child 0, iteration: int64
      child 1, turn: int64
      child 2, accumulated_directives: list<item: string>
          child 0, item: string
      child 3, solver_prompt: string
      child 4, solver_output: string
      child 5, solver_validation: struct<has_answer_sentinel: bool, extracted_candidate_answer: string, extracted_reference_answer: st (... 143 chars omitted)
          child 0, has_answer_sentinel: bool
          child 1, extracted_candidate_answer: string
          child 2, extracted_reference_answer: string
          child 3, is_equivalent: bool
          child 4, match_method: string
          child 5, relative_error: double
          child 6, unit_consistent: bool
          child 7, candidate_unit: string
          child 8, reference_unit: string
      child 6, solver_is_correct: bool
      child 7, solver_tokens: int64
      child 8, solver_latency_s: double
      child 9, verifier_prompt: string
      child 10, verifier_output: string
      child 11, verifier_verdict: string
      child 12, verifier_directive: string
      child 13, verifier_is_decision_correct: bool
      child 14, verifier_tokens: int64
      child 15, verifier_latency_s: double
problem_id: string
turn0_correct: bool
final_has_answer: bool
converted: bool
total_latency_s: double
problem_statement: string
regressed: bool
domain: string
final_correct: bool
reference_solution: string
to
{'problem_id': Value('string'), 'domain': Value('string'), 'problem_statement': Value('string'), 'reference_solution': Value('string'), 'normalized_reference_answer': Value('string'), 'iterations_used': Value('int64'), 'max_iterations': Value('int64'), 'stop_reason': Value('string'), 'turn0_correct': Value('bool'), 'final_correct': Value('bool'), 'turn0_has_answer': Value('bool'), 'final_has_answer': Value('bool'), 'turn0_match_method': Value('string'), 'final_match_method': Value('string'), 'converted': Value('bool'), 'regressed': Value('bool'), 'total_latency_s': Value('float64'), 'total_tokens': Value('int64'), 'traces': List({'iteration': Value('int64'), 'turn': Value('int64'), 'accumulated_directives': List(Value('string')), 'solver_prompt': Value('string'), 'solver_output': Value('string'), 'solver_validation': {'has_answer_sentinel': Value('bool'), 'extracted_candidate_answer': Value('string'), 'extracted_reference_answer': Value('string'), 'is_equivalent': Value('bool'), 'match_method': Value('string'), 'relative_error': Value('float64'), 'unit_consistent': Value('bool'), 'candidate_unit': Value('string'), 'reference_unit': Value('string')}, 'solver_is_correct': Value('bool'), 'solver_tokens': Value('int64'), 'solver_latency_s': Value('float64'), 'verifier_prompt': Value('string'), 'verifier_output': Value('string'), 'verifier_verdict': Value('string'), 'verifier_directive': Value('string'), 'verifier_is_decision_correct': Value('bool'), 'verifier_tokens': Value('int64'), 'verifier_latency_s': Value('float64')})}
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 544, 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 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 361, 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 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              metadata: struct<validation_source: string, validation_local_path: string, limit: null, batch_size: int64, num (... 266 chars omitted)
                child 0, validation_source: string
                child 1, validation_local_path: string
                child 2, limit: null
                child 3, batch_size: int64
                child 4, num_gpu_workers: int64
                child 5, tier: string
                child 6, curated_first: bool
                child 7, base_model: string
                child 8, solver_adapter: string
                child 9, verifier_adapter: string
                child 10, prompt_mode: string
                child 11, solver_max_new_tokens: int64
                child 12, verifier_max_new_tokens: int64
                child 13, temperature_solver: double
                child 14, temperature_verifier: double
              metrics: struct<total_problems: int64, max_iterations: int64, solver: struct<turn0_correct: int64, turn0_accu (... 3067 chars omitted)
                child 0, total_problems: int64
                child 1, max_iterations: int64
                child 2, solver: struct<turn0_correct: int64, turn0_accuracy: double, final_correct: int64, final_accuracy: double, n (... 566 chars omitted)
                    child 0, turn0_correct: int64
                    child 1, turn0_accuracy: double
                    child 2, final_correct: int64
                    child 3, final_accuracy: double
                    child 4, net_revision_gain: double
                    child 5, turn0_sentinel_compliance_rate: double
                    child 6, final_sentinel_compliance_rate: double
                    child 7, converted_count: int64
                    child 8, turn0_wrong_count: int64
                    child 9, correction_conversion_rate: double
                    child 10, regressed_count: int64
                    child 11, regression_rate: double
               
              ...
              4, turn: int64, accumulated_directives: list<item: string>, solver_prompt: str (... 556 chars omitted)
                    child 0, iteration: int64
                    child 1, turn: int64
                    child 2, accumulated_directives: list<item: string>
                        child 0, item: string
                    child 3, solver_prompt: string
                    child 4, solver_output: string
                    child 5, solver_validation: struct<has_answer_sentinel: bool, extracted_candidate_answer: string, extracted_reference_answer: st (... 143 chars omitted)
                        child 0, has_answer_sentinel: bool
                        child 1, extracted_candidate_answer: string
                        child 2, extracted_reference_answer: string
                        child 3, is_equivalent: bool
                        child 4, match_method: string
                        child 5, relative_error: double
                        child 6, unit_consistent: bool
                        child 7, candidate_unit: string
                        child 8, reference_unit: string
                    child 6, solver_is_correct: bool
                    child 7, solver_tokens: int64
                    child 8, solver_latency_s: double
                    child 9, verifier_prompt: string
                    child 10, verifier_output: string
                    child 11, verifier_verdict: string
                    child 12, verifier_directive: string
                    child 13, verifier_is_decision_correct: bool
                    child 14, verifier_tokens: int64
                    child 15, verifier_latency_s: double
              problem_id: string
              turn0_correct: bool
              final_has_answer: bool
              converted: bool
              total_latency_s: double
              problem_statement: string
              regressed: bool
              domain: string
              final_correct: bool
              reference_solution: string
              to
              {'problem_id': Value('string'), 'domain': Value('string'), 'problem_statement': Value('string'), 'reference_solution': Value('string'), 'normalized_reference_answer': Value('string'), 'iterations_used': Value('int64'), 'max_iterations': Value('int64'), 'stop_reason': Value('string'), 'turn0_correct': Value('bool'), 'final_correct': Value('bool'), 'turn0_has_answer': Value('bool'), 'final_has_answer': Value('bool'), 'turn0_match_method': Value('string'), 'final_match_method': Value('string'), 'converted': Value('bool'), 'regressed': Value('bool'), 'total_latency_s': Value('float64'), 'total_tokens': Value('int64'), 'traces': List({'iteration': Value('int64'), 'turn': Value('int64'), 'accumulated_directives': List(Value('string')), 'solver_prompt': Value('string'), 'solver_output': Value('string'), 'solver_validation': {'has_answer_sentinel': Value('bool'), 'extracted_candidate_answer': Value('string'), 'extracted_reference_answer': Value('string'), 'is_equivalent': Value('bool'), 'match_method': Value('string'), 'relative_error': Value('float64'), 'unit_consistent': Value('bool'), 'candidate_unit': Value('string'), 'reference_unit': Value('string')}, 'solver_is_correct': Value('bool'), 'solver_tokens': Value('int64'), 'solver_latency_s': Value('float64'), 'verifier_prompt': Value('string'), 'verifier_output': Value('string'), 'verifier_verdict': Value('string'), 'verifier_directive': Value('string'), 'verifier_is_decision_correct': Value('bool'), 'verifier_tokens': Value('int64'), 'verifier_latency_s': Value('float64')})}
              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.

Adversarial Physics Solver-Verifier — Benchmarks, Multi-Turn Traces & Audit Reports

This repository complements the 15 training datasets (vital-ar/physics-*) and 15 LoRA adapters (vital-ar/adversarial-physics-*) from our Lexsi AI Safety & Alignment Hackathon (Track 1, Problem Statement 8: Data-Scarce Domains) submission.

Contents

  • splits/ — Canonical PhysGym seed split (train_seeds.jsonl, 21 rows), plus all-MiniLM-L6-v2 cosine < 0.80 leak-purged evaluation splits:
    • v1: validation.jsonl (38 rows), test.jsonl (37 rows — pruned 1 >= 0.80 item physgym_0204)
    • v2: validation_v2.jsonl (44 rows), test_v2.jsonl (49 rows — pruned 16 val / 11 test)
    • v3: validation_v3.jsonl (48 rows), test_v3.jsonl (55 rows — pruned 12 val / 5 test)
  • traces/eval_e4_loop_*/ — Full raw multi-turn Solver–Verifier (E4a + E4b) interaction traces (results_per_problem.jsonl and summary.json) for v2_test, v3_val, v3_test, plus the pre-fix verifier prefix-leak ablation (traces/ablations/eval_e4_loop_v3_prefix_leaked/summary.json).
  • traces/single_turn/ — Raw per-problem single-turn AuditKIT + SymPy evaluation JSONs across v1, v2, and v3 (E0–E4a).
  • reports/ — Dual Table 1 (Original Evaluated Splits) & Table 2 (Clean cos < 0.80 Recalculated Splits) reports:
    • comparison_tables/experiment_comparison_v1_v2_v3.{md,json,csv} and experiment_comparison_v1_v2_v3_clean_080.csv
    • audit_reports/leak_and_calibration_summary.json (live < 0.80 semantic leak audits, pre-filter archive baseline, Verifier Orig vs Clean Cohen's kappa, SymPy vs AlignTune kappa, and compute ledger)
    • audit_reports/excluded_leaked_eval_items_080.json (all archived raw >= 0.80 items and closest training matches)

License & Provenance

  • Held-out PhysGym problems (validation.jsonl, test.jsonl, train_seeds.jsonl) originate from YimengChen/PhysGym (CC0-1.0).
  • Held-out curated textbook problems originate from Benjamin Crowell's Light and Matter / Simple Nature series (CC BY-SA 4.0).
  • Evaluation traces, comparison tables, and calibration reports are released under CC BY-SA 4.0.
Downloads last month
-
Total file size:
10.5 MB

Collection including vital-ar/adversarial-physics-eval-traces-benchmarks