The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 80, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.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.
Chronos Policy: simulation data and evaluated simulators
All eight prepared training stores and both evaluated simulator snapshots are publicly available. The Conveyor/Pivoting training and evaluation campaign is complete. The optional paper reproduction workflow is at https://github.com/changhaowang/ilp/tree/release. The GitHub repository currently requires access; public code publication is pending.
Evaluated simulator snapshots
simulators/manifest.json lists immutable, checksum-verified bundles containing
the evaluated Conveyor/DexMimicGen and floating-book Pivoting simulator snapshots.
Each archive includes a BUNDLE_MANIFEST.json with per-file SHA-256 hashes.
Original third-party licenses and notices are preserved inside the bundles.
The bundles contain source and physics assets. Install the Python dependencies following the code release's Conveyor/Pivoting and DexMimicGen setup instructions. They are not standalone Python environments.
The Conveyor snapshot preserves the historical paper reset behavior. A diagnostic confirmed that its reset state can depend on the preceding trajectory. Fixed scene seeds alone therefore do not guarantee identical historical scores. This behavior is being investigated; it has not been silently changed in the paper replay.
Prepared training datasets
The following prepared training stores are uploaded and checksum verified:
| Task | Episodes | Download | Extracted files |
|---|---|---|---|
Threading (threading) |
827 | 196.8 GB | 269.4 GB |
Assembly (assembly) |
830 | 240.8 GB | 337.4 GB |
Transport (transport) |
1,029 | 71.1 GB | 116.8 GB |
Coffee (coffee) |
1,014 | 26.1 GB | 39.8 GB |
Drawer Cleanup (drawer_cleanup) |
1,026 | 23.9 GB | 34.1 GB |
Pouring (pouring) |
1,009 | 28.4 GB | 42.1 GB |
Conveyor (conveyor) |
1,742 | 53.9 GB | 74.9 GB |
Pivoting (flipup) |
270 | 11.5 GB | 13.0 GB |
Keeping all archives and extracted stores requires approximately 1.6 TB, plus space for checkpoints and training outputs. Download only the tasks you need.
After setting up the code release, use its verified downloader. These public artifacts do not require a Hugging Face login:
python -m chronos.reproduce artifacts --task conveyor --kind data
python -m chronos.reproduce artifacts --task flipup --kind data
python -m chronos.reproduce artifacts --task threading --kind data
python -m chronos.reproduce artifacts --task coffee --kind data
python -m chronos.reproduce artifacts --task pouring --kind data
python -m chronos.reproduce artifacts --task conveyor --kind simulators
The commands use immutable revisions and verify every extracted file before
installation. See the release's docs/reproduction.md for setup and disk paths.
Each available task has
a datasets/<task>/manifest.json recording every archive's size and SHA-256.
files.jsonl.gz records the hash of every original Zarr file, and every tar member
is verified against it before upload. Extract all of a task's part-*.tar.gz
archives into one data/ directory to obtain data/<task>/episode_data.zarr.
All eight task manifests are included in the pinned release revision.
DexMimicGen source data are from https://huggingface.co/datasets/MimicGen/dexmimicgen_datasets and retain their CC BY-NC-SA 4.0 terms and attribution. Threading and Assembly use prepared 256-pixel images; Transport, Coffee, Drawer Cleanup, and Pouring use prepared 84-pixel images resized by the recorded model preprocessing. All use the recorded 100 Hz prepared stores and absolute-action conventions. The original HDF5 environment metadata can be obtained from the upstream dataset repository.
Original paper weights: https://huggingface.co/changhaowang/chronos-policy
- Downloads last month
- 121