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 84, 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.
Two complete pretraining folders
Download one complete archive for each independently usable pretraining dataset:
- pretraining_linker.tar.gz: 39,016 records: 10,255 2D heavy-atom graphs and 28,761 computed 3D conformers.
- pretraining_water.tar.gz: 15,594 local protein-ligand-observed-water scenes.
Each archive contains the entire corresponding prepared training folder, including all NPZ samples, per-parent record.json files, manifest.jsonl, normalization.json, summary.json, and fixed train/validation/test indexes. No additional partition archive is required. The original file contents and split assignments are preserved.
Download with the web links above or the official CLI:
hf download cpy456/protac-water-pretraining --repo-type dataset --local-dir pretraining_download
cd pretraining_download
sha256sum -c checksums.sha256
Extract from your project root (substitute the downloaded archive location):
tar -xzf pretraining_download/pretraining_linker.tar.gz
tar -xzf pretraining_download/pretraining_water.tar.gz
This restores output/pretraining_linker/ and output/pretraining_water/.
Paths in the manifests remain relative to each restored training folder.
Private repositories require an authorized Hugging Face login for download.
Public repositories can be downloaded without login.
See DATA_FORMAT.html and dataset_info.json for field shapes and real examples. NPZ files are read using numpy.load(path, allow_pickle=False). The two pretraining branches are independent datasets, not paired PROTAC/water joint-training examples. Within each water scene the environment and observed water are aligned; 3D conformers share their 2D chemical parent within one split.
Only active train/validation/test records are distributed. Reserved target data and quarantined observations are excluded. Training statistics use training partitions only. Linker splits isolate chemical groups; water splits isolate protein groups. Ligands can recur across the primary water partitions. Double-unseen water subsets are marked in the manifests.
Computed conformers are not experimental ternary poses. Water labels are observed oxygen positions, not all solvent, hydrogen orientations, or proven empty sites. These are derived tensors from the project's archived PROTAC chemistry/computed conformer and PDB-REDO sources. Upstream terms continue to apply; this card does not grant a new license. Provenance hashes are retained in record.json. Raw downloads, credentials, chat history, project memory, training logs and model weights are not included.
- Downloads last month
- -