Hugging Face's logo

Datasets:
rycerzes
/
fswe-pg-01-traj-q36-27b

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
episode_id: int64
reward: double
frozen_scores: struct<S1: double, S2: double, S3: double, S4: double, S5: double>
  child 0, S1: double
  child 1, S2: double
  child 2, S3: double
  child 3, S4: double
  child 4, S5: double
dense_rewards_used: bool
num_steps: int64
num_subtasks_covered: int64
subtask_reward_range: list<item: double>
  child 0, item: double
steps: list<item: struct<mean_logprob: double, action_token_count: int64, total_action_tokens: int64, skipp (... 216 chars omitted)
  child 0, item: struct<mean_logprob: double, action_token_count: int64, total_action_tokens: int64, skipped_action_t (... 204 chars omitted)
      child 0, mean_logprob: double
      child 1, action_token_count: int64
      child 2, total_action_tokens: int64
      child 3, skipped_action_tokens: int64
      child 4, logprob_start_len: int64
      child 5, step_index: int64
      child 6, message_index: int64
      child 7, subtask_id: string
      child 8, subtask_reward: double
      child 9, phase: string
      child 10, pi_hind: double
      child 11, rho: double
      child 12, q_h: double
      child 13, q_h_smoothed: double
pi_hind_mean: double
hyperparams: struct<t_temp: double, gamma: double, c_min: double, c_max: double, alpha: double, smooth: bool, den (... 44 chars omitted)
  child 0, t_temp: double
  child 1, gamma: double
  child 2, c_min: double
  child 3, c_max: double
  child 4, alpha: double
  child 5, smooth: bool
  child 6, dense_rewards: bool
  child 7, max_logprob_tokens: int64
total_episodes: int64
failed_episodes: int64
successful_episodes: int64
reward_stats: struct<min: double, max: double, mean: double, median: double, top_quartile_min: double, bottom_quar (... 17 chars omitted)
  child 0, min: double
  child 1, max: double
  child 2, mean: double
  child 3, median: double
  child 4, top_quartile_min: double
  child 5, bottom_quartile_max: double
episodes: list<item: struct<episode_id: int64, reward: double, phase: string, turns: int64, elapsed_s: double, (... 31 chars omitted)
  child 0, item: struct<episode_id: int64, reward: double, phase: string, turns: int64, elapsed_s: double, has_jsonl: (... 19 chars omitted)
      child 0, episode_id: int64
      child 1, reward: double
      child 2, phase: string
      child 3, turns: int64
      child 4, elapsed_s: double
      child 5, has_jsonl: bool
      child 6, error: null
elapsed_s: double
elapsed_min: double
to
{'total_episodes': Value('int64'), 'elapsed_s': Value('float64'), 'elapsed_min': Value('float64'), 'episodes': List({'episode_id': Value('int64'), 'reward': Value('float64'), 'phase': Value('string'), 'turns': Value('int64'), 'elapsed_s': Value('float64'), 'has_jsonl': Value('bool'), 'error': Value('null')}), 'successful_episodes': Value('int64'), 'failed_episodes': Value('int64'), 'reward_stats': {'min': Value('float64'), 'max': Value('float64'), 'mean': Value('float64'), 'median': Value('float64'), 'top_quartile_min': Value('float64'), 'bottom_quartile_max': Value('float64')}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
                  return get_rows(
                         ^^^^^^^^^
                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 77, in get_rows
                  rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2690, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2227, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2251, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 494, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 384, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 295, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 128, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2281, in table_cast
                  return cast_table_to_schema(table, schema)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2227, in cast_table_to_schema
                  raise CastError(
              datasets.table.CastError: Couldn't cast
              episode_id: int64
              reward: double
              frozen_scores: struct<S1: double, S2: double, S3: double, S4: double, S5: double>
                child 0, S1: double
                child 1, S2: double
                child 2, S3: double
                child 3, S4: double
                child 4, S5: double
              dense_rewards_used: bool
              num_steps: int64
              num_subtasks_covered: int64
              subtask_reward_range: list<item: double>
                child 0, item: double
              steps: list<item: struct<mean_logprob: double, action_token_count: int64, total_action_tokens: int64, skipp (... 216 chars omitted)
                child 0, item: struct<mean_logprob: double, action_token_count: int64, total_action_tokens: int64, skipped_action_t (... 204 chars omitted)
                    child 0, mean_logprob: double
                    child 1, action_token_count: int64
                    child 2, total_action_tokens: int64
                    child 3, skipped_action_tokens: int64
                    child 4, logprob_start_len: int64
                    child 5, step_index: int64
                    child 6, message_index: int64
                    child 7, subtask_id: string
                    child 8, subtask_reward: double
                    child 9, phase: string
                    child 10, pi_hind: double
                    child 11, rho: double
                    child 12, q_h: double
                    child 13, q_h_smoothed: double
              pi_hind_mean: double
              hyperparams: struct<t_temp: double, gamma: double, c_min: double, c_max: double, alpha: double, smooth: bool, den (... 44 chars omitted)
                child 0, t_temp: double
                child 1, gamma: double
                child 2, c_min: double
                child 3, c_max: double
                child 4, alpha: double
                child 5, smooth: bool
                child 6, dense_rewards: bool
                child 7, max_logprob_tokens: int64
              total_episodes: int64
              failed_episodes: int64
              successful_episodes: int64
              reward_stats: struct<min: double, max: double, mean: double, median: double, top_quartile_min: double, bottom_quar (... 17 chars omitted)
                child 0, min: double
                child 1, max: double
                child 2, mean: double
                child 3, median: double
                child 4, top_quartile_min: double
                child 5, bottom_quartile_max: double
              episodes: list<item: struct<episode_id: int64, reward: double, phase: string, turns: int64, elapsed_s: double, (... 31 chars omitted)
                child 0, item: struct<episode_id: int64, reward: double, phase: string, turns: int64, elapsed_s: double, has_jsonl: (... 19 chars omitted)
                    child 0, episode_id: int64
                    child 1, reward: double
                    child 2, phase: string
                    child 3, turns: int64
                    child 4, elapsed_s: double
                    child 5, has_jsonl: bool
                    child 6, error: null
              elapsed_s: double
              elapsed_min: double
              to
              {'total_episodes': Value('int64'), 'elapsed_s': Value('float64'), 'elapsed_min': Value('float64'), 'episodes': List({'episode_id': Value('int64'), 'reward': Value('float64'), 'phase': Value('string'), 'turns': Value('int64'), 'elapsed_s': Value('float64'), 'has_jsonl': Value('bool'), 'error': Value('null')}), 'successful_episodes': Value('int64'), 'failed_episodes': Value('int64'), 'reward_stats': {'min': Value('float64'), 'max': Value('float64'), 'mean': Value('float64'), 'median': Value('float64'), 'top_quartile_min': Value('float64'), 'bottom_quartile_max': 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.

fswe-pg-01 trajectory episodes

This dataset holds raw episode trajectories for the Frontier SWE (FSWE) postgres task: implement a PostgreSQL wire protocol server that fronts SQLite (see task config in the upstream frontier-swe-openenv repo under frontier_swe_env/tasks/pg.py). This run is referred to as pg-01 in tooling and job names.

Model

Rollouts and agent behavior were produced with Qwen/Qwen3.6-27B (Qwen 3.6 27B). Hindsight scoring (compute_hindsight_scores.py) uses the same family of stack configured for that script (e.g. SGLang /generate for log-probabilities).

How the data was produced

  1. collect_trajectories.py — Runs parallel Docker workers, one Frontier SWE episode per episode_NNN/ directory (NNN is zero-padded). After each episode it writes result.json, then pulls pi_session.jsonl from the Pi session path inside the container (docker exec + find + docker cp) and container_logs.txt from docker logs (stdout + stderr).
  2. backfill_rewards.py — Batch pass to fill missing episode_reward (and related fields) when episodes ended without a server-side reward.
  3. compute_hindsight_scores.py — For each assistant step, builds a hindsight prompt with the final outcome (reward, phase, subtask scores), queries the inference API for log-probs of the original action tokens, and computes HCAPO hindsight Q_H / importance ratios (paper 2603.08754). Writes hindsight_scores.json per episode.

Layout 

trajectories/
  episode_000/
    result.json
    pi_session.jsonl
    container_logs.txt
    hindsight_scores.json   # present after compute_hindsight_scores
  episode_001/
    ...
File Role
result.json Episode summary: turns, phase, scores, episode_reward, metadata.
pi_session.jsonl Full Pi agent trace (tool calls, model output).
container_logs.txt Combined container stdout/stderr for debugging / scoring.
hindsight_scores.json Per-step hindsight Q values for HCAPO training (not from collection alone).
Downloads last month
78
Total file size:
12.8 MB

Article mentioning rycerzes/fswe-pg-01-traj-q36-27b