Datasets:

ccwatson
/

metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
accepted_episode_count: int64
total_frames: int64
task_count: int64
episodes_per_task_variant: struct<red-to-red/avoid-disk-00: int64, red-to-red/no-avoid-disk-00: int64, blue-to-blue/avoid-disk- (... 48 chars omitted)
  child 0, red-to-red/avoid-disk-00: int64
  child 1, red-to-red/no-avoid-disk-00: int64
  child 2, blue-to-blue/avoid-disk-00: int64
  child 3, blue-to-blue/no-avoid-disk-00: int64
quality_flag_count: int64
generation_elapsed_s: double
generation_episodes_per_min: double
merge_elapsed_s: double
action_noise: double
randomization: struct<block_start_xy_noise_magnitude_m: double, target_xy_noise_magnitude_m: double>
  child 0, block_start_xy_noise_magnitude_m: double
  child 1, target_xy_noise_magnitude_m: double
task_summary: list<item: struct<task_index: int64, task_key: string, prompt: string, programmatic_prompt: string,  (... 602 chars omitted)
  child 0, item: struct<task_index: int64, task_key: string, prompt: string, programmatic_prompt: string, target_bloc (... 590 chars omitted)
      child 0, task_index: int64
      child 1, task_key: string
      child 2, prompt: string
      child 3, programmatic_prompt: string
      child 4, target_block_color: string
      child 5, destination_bin_color: string
      child 6, avoid_disk_obstacle: bool
      child 7, accepted_episodes: int64
      child 8, frames: int64
      child 9, final_validation_failures: int64
      child 10, steps: struct<count: int64, min: double, max: double, mean: double, median: double,
...
, worker_count: int64, shards_per_task: int64, episodes_per_shard: i (... 483 chars omitted)
  child 0, parallel_strategy: string
  child 1, worker_count: int64
  child 2, shards_per_task: int64
  child 3, episodes_per_shard: int64
  child 4, generation_elapsed_s: double
  child 5, generation_episodes_per_min: double
  child 6, shard_generation_summary_path: string
  child 7, shard_generation_commands_path: string
  child 8, exact_commands: struct<run_name: string, hub_repo_id: string, started_at_utc: timestamp[s], worker_count: int64, epi (... 214 chars omitted)
      child 0, run_name: string
      child 1, hub_repo_id: string
      child 2, started_at_utc: timestamp[s]
      child 3, worker_count: int64
      child 4, episodes_per_shard: int64
      child 5, shards_per_task: int64
      child 6, task_keys: list<item: string>
          child 0, item: string
      child 7, commands: list<item: struct<shard_index: int64, task_key: string, seed: int64, output_root: string, logfile: s (... 24 chars omitted)
          child 0, item: struct<shard_index: int64, task_key: string, seed: int64, output_root: string, logfile: string, comm (... 12 chars omitted)
              child 0, shard_index: int64
              child 1, task_key: string
              child 2, seed: int64
              child 3, output_root: string
              child 4, logfile: string
              child 5, command: string
merge_started_at_utc: timestamp[s]
dataset_name: string
schema_version: int64
format: string
to
{'schema_version': Value('int64'), 'dataset_name': Value('string'), 'hub_repo_id': Value('string'), 'written_at_utc': Value('timestamp[s]'), 'merge_started_at_utc': Value('timestamp[s]'), 'merge_elapsed_s': Value('float64'), 'provenance': {'git': {'repo_root': Value('string'), 'commit': Value('string'), 'commit_short': Value('string'), 'branch': Value('string'), 'remote': Value('string'), 'dirty': Value('bool'), 'dirty_file_count': Value('int64'), 'dirty_files': List(Value('string'))}, 'command_line': {'argv': List(Value('string')), 'command': Value('string'), 'python_command': Value('string'), 'launch_command': Value('null'), 'launch_script': Value('null'), 'cwd': Value('string'), 'executable': Value('string')}, 'env': {}, 'slurm': {'running_under_slurm': Value('bool'), 'env': {}}}, 'intention': Value('string'), 'format': Value('string'), 'generation': {'parallel_strategy': Value('string'), 'worker_count': Value('int64'), 'shards_per_task': Value('int64'), 'episodes_per_shard': Value('int64'), 'generation_elapsed_s': Value('float64'), 'generation_episodes_per_min': Value('float64'), 'shard_generation_summary_path': Value('string'), 'shard_generation_commands_path': Value('string'), 'exact_commands': {'run_name': Value('string'), 'hub_repo_id': Value('string'), 'started_at_utc': Value('timestamp[s]'), 'worker_count': Value('int64'), 'episodes_per_shard': Value('int64'), 'shards_per_task': Value('int64'), 'task_keys': List(Value('string')), 'commands': List({'shard_index': Val
...
float64'), 'max': Value('float64'), 'mean': Value('float64'), 'median': Value('float64'), 'std': Value('float64')}, 'min_disk_center_distance_m': {'count': Value('int64'), 'min': Value('float64'), 'max': Value('float64'), 'mean': Value('float64'), 'median': Value('float64'), 'std': Value('float64')}, 'max_edge_clearance_m': {'count': Value('int64'), 'min': Value('float64'), 'max': Value('float64'), 'mean': Value('float64'), 'median': Value('float64'), 'std': Value('float64')}}), 'validation': {'quality_flag_count': Value('int64'), 'success_episodes': Value('int64'), 'disk_rule': Value('string')}, 'throughput_probe': {'no_video_eval': List({'workers': Value('int64'), 'episodes_requested': Value('int64'), 'episodes_completed': Value('int64'), 'elapsed_s': Value('float64'), 'episodes_per_s': Value('float64'), 'episodes_per_min': Value('float64'), 'peak_gpu_mem_used_mib': Value('int64'), 'peak_gpu_util_percent': Value('int64'), 'failures': List(Value('null'))}), 'writer': List({'workers': Value('int64'), 'episodes_completed': Value('int64'), 'elapsed_s': Value('float64'), 'episodes_per_s': Value('float64'), 'episodes_per_min': Value('float64'), 'peak_gpu_mem_used_mib': Value('int64'), 'peak_gpu_util_percent': Value('int64'), 'bytes_written': Value('int64'), 'mib_per_episode': Value('float64'), 'failures': List(Value('null'))})}, 'paths': {'dataset_root': Value('string'), 'artifact_root': Value('string'), 'source_shards_root': Value('string'), 'run_metadata_dir': 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(
                         ^^^^^^^^^
                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 478, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2815, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2352, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2377, 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 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.12/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/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 310, 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 130, 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 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
              datasets.table.CastError: Couldn't cast
              accepted_episode_count: int64
              total_frames: int64
              task_count: int64
              episodes_per_task_variant: struct<red-to-red/avoid-disk-00: int64, red-to-red/no-avoid-disk-00: int64, blue-to-blue/avoid-disk- (... 48 chars omitted)
                child 0, red-to-red/avoid-disk-00: int64
                child 1, red-to-red/no-avoid-disk-00: int64
                child 2, blue-to-blue/avoid-disk-00: int64
                child 3, blue-to-blue/no-avoid-disk-00: int64
              quality_flag_count: int64
              generation_elapsed_s: double
              generation_episodes_per_min: double
              merge_elapsed_s: double
              action_noise: double
              randomization: struct<block_start_xy_noise_magnitude_m: double, target_xy_noise_magnitude_m: double>
                child 0, block_start_xy_noise_magnitude_m: double
                child 1, target_xy_noise_magnitude_m: double
              task_summary: list<item: struct<task_index: int64, task_key: string, prompt: string, programmatic_prompt: string,  (... 602 chars omitted)
                child 0, item: struct<task_index: int64, task_key: string, prompt: string, programmatic_prompt: string, target_bloc (... 590 chars omitted)
                    child 0, task_index: int64
                    child 1, task_key: string
                    child 2, prompt: string
                    child 3, programmatic_prompt: string
                    child 4, target_block_color: string
                    child 5, destination_bin_color: string
                    child 6, avoid_disk_obstacle: bool
                    child 7, accepted_episodes: int64
                    child 8, frames: int64
                    child 9, final_validation_failures: int64
                    child 10, steps: struct<count: int64, min: double, max: double, mean: double, median: double,
              ...
              , worker_count: int64, shards_per_task: int64, episodes_per_shard: i (... 483 chars omitted)
                child 0, parallel_strategy: string
                child 1, worker_count: int64
                child 2, shards_per_task: int64
                child 3, episodes_per_shard: int64
                child 4, generation_elapsed_s: double
                child 5, generation_episodes_per_min: double
                child 6, shard_generation_summary_path: string
                child 7, shard_generation_commands_path: string
                child 8, exact_commands: struct<run_name: string, hub_repo_id: string, started_at_utc: timestamp[s], worker_count: int64, epi (... 214 chars omitted)
                    child 0, run_name: string
                    child 1, hub_repo_id: string
                    child 2, started_at_utc: timestamp[s]
                    child 3, worker_count: int64
                    child 4, episodes_per_shard: int64
                    child 5, shards_per_task: int64
                    child 6, task_keys: list<item: string>
                        child 0, item: string
                    child 7, commands: list<item: struct<shard_index: int64, task_key: string, seed: int64, output_root: string, logfile: s (... 24 chars omitted)
                        child 0, item: struct<shard_index: int64, task_key: string, seed: int64, output_root: string, logfile: string, comm (... 12 chars omitted)
                            child 0, shard_index: int64
                            child 1, task_key: string
                            child 2, seed: int64
                            child 3, output_root: string
                            child 4, logfile: string
                            child 5, command: string
              merge_started_at_utc: timestamp[s]
              dataset_name: string
              schema_version: int64
              format: string
              to
              {'schema_version': Value('int64'), 'dataset_name': Value('string'), 'hub_repo_id': Value('string'), 'written_at_utc': Value('timestamp[s]'), 'merge_started_at_utc': Value('timestamp[s]'), 'merge_elapsed_s': Value('float64'), 'provenance': {'git': {'repo_root': Value('string'), 'commit': Value('string'), 'commit_short': Value('string'), 'branch': Value('string'), 'remote': Value('string'), 'dirty': Value('bool'), 'dirty_file_count': Value('int64'), 'dirty_files': List(Value('string'))}, 'command_line': {'argv': List(Value('string')), 'command': Value('string'), 'python_command': Value('string'), 'launch_command': Value('null'), 'launch_script': Value('null'), 'cwd': Value('string'), 'executable': Value('string')}, 'env': {}, 'slurm': {'running_under_slurm': Value('bool'), 'env': {}}}, 'intention': Value('string'), 'format': Value('string'), 'generation': {'parallel_strategy': Value('string'), 'worker_count': Value('int64'), 'shards_per_task': Value('int64'), 'episodes_per_shard': Value('int64'), 'generation_elapsed_s': Value('float64'), 'generation_episodes_per_min': Value('float64'), 'shard_generation_summary_path': Value('string'), 'shard_generation_commands_path': Value('string'), 'exact_commands': {'run_name': Value('string'), 'hub_repo_id': Value('string'), 'started_at_utc': Value('timestamp[s]'), 'worker_count': Value('int64'), 'episodes_per_shard': Value('int64'), 'shards_per_task': Value('int64'), 'task_keys': List(Value('string')), 'commands': List({'shard_index': Val
              ...
              float64'), 'max': Value('float64'), 'mean': Value('float64'), 'median': Value('float64'), 'std': Value('float64')}, 'min_disk_center_distance_m': {'count': Value('int64'), 'min': Value('float64'), 'max': Value('float64'), 'mean': Value('float64'), 'median': Value('float64'), 'std': Value('float64')}, 'max_edge_clearance_m': {'count': Value('int64'), 'min': Value('float64'), 'max': Value('float64'), 'mean': Value('float64'), 'median': Value('float64'), 'std': Value('float64')}}), 'validation': {'quality_flag_count': Value('int64'), 'success_episodes': Value('int64'), 'disk_rule': Value('string')}, 'throughput_probe': {'no_video_eval': List({'workers': Value('int64'), 'episodes_requested': Value('int64'), 'episodes_completed': Value('int64'), 'elapsed_s': Value('float64'), 'episodes_per_s': Value('float64'), 'episodes_per_min': Value('float64'), 'peak_gpu_mem_used_mib': Value('int64'), 'peak_gpu_util_percent': Value('int64'), 'failures': List(Value('null'))}), 'writer': List({'workers': Value('int64'), 'episodes_completed': Value('int64'), 'elapsed_s': Value('float64'), 'episodes_per_s': Value('float64'), 'episodes_per_min': Value('float64'), 'peak_gpu_mem_used_mib': Value('int64'), 'peak_gpu_util_percent': Value('int64'), 'bytes_written': Value('int64'), 'mib_per_episode': Value('float64'), 'failures': List(Value('null'))})}, 'paths': {'dataset_root': Value('string'), 'artifact_root': Value('string'), 'source_shards_root': Value('string'), 'run_metadata_dir': 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.

MetaWorld Bin Transfer Disk Obstacle Randomized Noise005 3200eps

This dataset contains scripted MetaWorld demonstrations for a two-bin colored-block transfer task with a visible green disk between the red and blue bins. The disk is an imaginary obstacle: in avoid-disk tasks the held block is not allowed above the disk, while in no-avoid tasks the same disk remains visible but the prompt explicitly says it does not need to be avoided.

High-Level Facts

HF repo: ccwatson/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps
Format: LeRobot v2.1 parquet episodes with embedded RGB image bytes.
Episodes: 3200
Frames: 545008
Task variants: 4, with 800 episodes each.
Cameras: corner4.image and gripperPOV.image, both 224 x 224 RGB.
State: observation.state shape [4]; observation.environment_state shape [39].
Action: actions shape [4]; records the clean scripted policy action.
Simulator action: clean policy action plus unrecorded clipped Gaussian noise on xyz dimensions, magnitude 0.05.
Start/target randomization: Gaussian xy noise with block_start_xy_noise_magnitude_m=0.008 and target_xy_noise_magnitude_m=0.012.
Disk: center [0.0, 0.82], radius 0.045 m, clearance 0.035 m, forbidden radius 0.08 m, arc points 7.
Generation: 16 parallel shard workers, four shards per task, 200 episodes per shard.
Shard generation time: 755.3 s (254.2 episodes/min).
Merge time: 113.5 s.
Quality flags: 0.

Task Variants

task_key	natural prompt	programmatic prompt	episodes	disk avoidance required
`red-to-red/avoid-disk-00`	move the red block to the red bin while avoiding the area above the green disk	`(:action transfer-block :parameters (red_block red_bin) :constraint (not-above red_block green_disk) :effect (in red_block red_bin))`	800	True
`red-to-red/no-avoid-disk-00`	move the red block to the red bin without needing to avoid the green disk	`(:action transfer-block :parameters (red_block red_bin) :context (disk-present green_disk) :effect (in red_block red_bin))`	800	False
`blue-to-blue/avoid-disk-00`	move the blue block to the blue bin while avoiding the area above the green disk	`(:action transfer-block :parameters (blue_block blue_bin) :constraint (not-above blue_block green_disk) :effect (in blue_block blue_bin))`	800	True
`blue-to-blue/no-avoid-disk-00`	move the blue block to the blue bin without needing to avoid the green disk	`(:action transfer-block :parameters (blue_block blue_bin) :context (disk-present green_disk) :effect (in blue_block blue_bin))`	800	False

Exact Generation Commands

The run used 16 shard commands. The complete machine-readable list is in meta/bin_transfer_shard_generation_commands.json. Example shard commands:

uv run python examples/metaworld/generate_bin_transfer_dataset.py --benchmark-tasks-path examples/metaworld/bin_transfer_disk_obstacle_benchmark_tasks.json --episodes-per-task 200 --max-attempts-per-task 260 --max-steps 360 --post-success-steps 8 --block-start-xy-noise-magnitude-m 0.008 --target-xy-noise-magnitude-m 0.012 --unrecorded-action-noise-magnitude 0.05 --overwrite --task-keys red-to-red/avoid-disk-00 --seed 73000 --output-root /home/christopher/Documents/openpi-finetune/openpi-metaworld/data/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps_shards/shard_00_red-to-red_avoid-disk-00 --dataset-repo-id local/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps_shard_00 --hub-repo-id ccwatson/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps
uv run python examples/metaworld/generate_bin_transfer_dataset.py --benchmark-tasks-path examples/metaworld/bin_transfer_disk_obstacle_benchmark_tasks.json --episodes-per-task 200 --max-attempts-per-task 260 --max-steps 360 --post-success-steps 8 --block-start-xy-noise-magnitude-m 0.008 --target-xy-noise-magnitude-m 0.012 --unrecorded-action-noise-magnitude 0.05 --overwrite --task-keys red-to-red/avoid-disk-00 --seed 74000 --output-root /home/christopher/Documents/openpi-finetune/openpi-metaworld/data/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps_shards/shard_01_red-to-red_avoid-disk-00 --dataset-repo-id local/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps_shard_01 --hub-repo-id ccwatson/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps
uv run python examples/metaworld/generate_bin_transfer_dataset.py --benchmark-tasks-path examples/metaworld/bin_transfer_disk_obstacle_benchmark_tasks.json --episodes-per-task 200 --max-attempts-per-task 260 --max-steps 360 --post-success-steps 8 --block-start-xy-noise-magnitude-m 0.008 --target-xy-noise-magnitude-m 0.012 --unrecorded-action-noise-magnitude 0.05 --overwrite --task-keys red-to-red/avoid-disk-00 --seed 75000 --output-root /home/christopher/Documents/openpi-finetune/openpi-metaworld/data/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps_shards/shard_02_red-to-red_avoid-disk-00 --dataset-repo-id local/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps_shard_02 --hub-repo-id ccwatson/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps
uv run python examples/metaworld/generate_bin_transfer_dataset.py --benchmark-tasks-path examples/metaworld/bin_transfer_disk_obstacle_benchmark_tasks.json --episodes-per-task 200 --max-attempts-per-task 260 --max-steps 360 --post-success-steps 8 --block-start-xy-noise-magnitude-m 0.008 --target-xy-noise-magnitude-m 0.012 --unrecorded-action-noise-magnitude 0.05 --overwrite --task-keys red-to-red/avoid-disk-00 --seed 76000 --output-root /home/christopher/Documents/openpi-finetune/openpi-metaworld/data/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps_shards/shard_03_red-to-red_avoid-disk-00 --dataset-repo-id local/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps_shard_03 --hub-repo-id ccwatson/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps

After shard generation, the final LeRobot root was created by rewriting each episode parquet's episode_index, global index, and task_index columns and copying per-episode summaries and rollout videos into this folder. The merge/upload metadata is in meta/bin_transfer_collection_metadata.json.

Parallel Capacity Probe

Before collection, no-video eval probes and writer probes were run to choose worker count. The final writer probe reached about 122.6 episodes/min at 16 workers with peak GPU memory 4718 MiB. Full probe outputs are stored in:

meta/bin_transfer_parallel_capacity_results.json
meta/bin_transfer_writer_capacity_results.json

Organization

data/chunk-*/episode_*.parquet: LeRobot episode data.
meta/info.json, meta/tasks.jsonl, meta/episodes.jsonl, meta/episodes_stats.jsonl: standard LeRobot metadata.
meta/bin_transfer_collection_metadata.json: full provenance, task config, commands, throughput probes, and validation semantics.
meta/bin_transfer_task_summary.json: per-task aggregate metrics.
meta/bin_transfer_episode_summaries_compact.jsonl: one compact row per episode.
meta/bin_transfer_hf_staged_upload_results_restart_from_chunk002.json: staged HF upload commit log for the successful restart.
meta/bin_transfer_upload_staged_to_hf.py: upload helper used for the staged/restart upload.
metadata/episode_summaries/: full per-episode rollout summaries.
rollout_videos/: per-episode MP4 rollout videos from the collection camera.
meta/bin_transfer_disk_obstacle_programmatic_prompt_remapping.yaml: natural-to-programmatic prompt mapping.

Intended Use

This dataset is intended for OpenPI / pi0.5 fine-tuning and prompt-format experiments around precise spatial constraints. The key contrast is whether the policy learns that the visible disk should be treated as a constraint only when the language asks for avoidance.

Exact Upload Commands And Recovery

The first upload attempt used upload_large_folder:

HF_HUB_DISABLE_XET=1 UV_CACHE_DIR=/tmp/openpi-uv-cache uv run python -c "from huggingface_hub import HfApi; HfApi().upload_large_folder(repo_id='ccwatson/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps', repo_type='dataset', folder_path='/home/christopher/Documents/openpi-finetune/openpi-metaworld/data/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps', private=False, num_workers=8, print_report=True, print_report_every=60)"

That path hit Hugging Face API rate limits (2500 api requests per 5 minutes) because the dataset contains thousands of small MP4/JSON files. The upload was then switched to a staged helper that commits bounded directory batches through HfApi.create_commit.

The staged upload command was:

HF_HUB_DISABLE_XET=1 HF_HUB_DISABLE_PROGRESS_BARS=1 UV_CACHE_DIR=/tmp/openpi-uv-cache uv run python outputs/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps/upload_staged_to_hf.py --repo-id ccwatson/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps --dataset-root data/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps --num-threads 2 --retry-wait-seconds 330 --max-retries 4 --results-json outputs/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps/hf_staged_upload_results.json

After data/chunk-000, data/chunk-001, and data/chunk-002 had reached the remote, the process was restarted to avoid a suspected HF stall and to split remaining commits into 250-file batches:

HF_HUB_DISABLE_XET=1 HF_HUB_DISABLE_PROGRESS_BARS=1 UV_CACHE_DIR=/tmp/openpi-uv-cache uv run python outputs/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps/upload_staged_to_hf.py --repo-id ccwatson/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps --dataset-root data/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps --start-stage data_chunk_002 --max-files-per-commit 250 --num-threads 2 --retry-wait-seconds 330 --max-retries 4 --results-json outputs/metaworld_bin_transfer_disk_obstacle_randomized_noise005_3200eps/hf_staged_upload_results_restart_from_chunk002.json

Remote verification after upload found:

total remote files listed by HfApi.list_repo_files: 9621
parquet episode files: 3200
rollout videos: 3200
full per-episode summary JSON files: 3200

Upload metadata is stored in meta/bin_transfer_upload_metadata.json, and the successful restart log is stored in meta/bin_transfer_hf_staged_upload_results_restart_from_chunk002.json.

Code Provenance

The exact git/command/environment provenance captured at merge time is stored in meta/bin_transfer_collection_metadata.json. The shard generator also wrote per-shard meta/bin_transfer_collection_metadata.json files in the source shard roots listed in meta/bin_transfer_source_episode_map.jsonl.

Downloads last month: 3,219

Total file size:

49.7 GB