Datasets:

yzhuang1997
/

chemvl-data

License:

Dataset Viewer

The dataset viewer is not available for this subset.

Cannot get the split names for the config 'default' of the dataset.

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 81, in _split_generators
                  first_examples = list(islice(pipeline, self.NUM_EXAMPLES_FOR_FEATURES_INFERENCE))
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 57, in _get_pipeline_from_tar
                  current_example[field_name] = cls.DECODERS[data_extension](current_example[field_name])
                                                ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 335, in torch_loads
                  return torch.load(io.BytesIO(data), weights_only=True)
                         ~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 1570, in load
                  return _load(
                      opened_zipfile,
                  ...<3 lines>...
                      **pickle_load_args,
                  )
                File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 2190, in _load
                  result = unpickler.load()
                File "/usr/local/lib/python3.14/site-packages/torch/_weights_only_unpickler.py", line 541, in load
                  self.append(self.persistent_load(pid))
                              ~~~~~~~~~~~~~~~~~~~~^^^^^
                File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 2154, in persistent_load
                  typed_storage = load_tensor(
                      dtype, nbytes, key, _maybe_decode_ascii(location)
                  )
                File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 2116, in load_tensor
                  wrap_storage = restore_location(storage, location)
                File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 734, in default_restore_location
                  result = fn(storage, location)
                File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 667, in _deserialize
                  device = _validate_device(location, backend_name)
                File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 634, in _validate_device
                  raise RuntimeError(
                  ...<5 lines>...
                  )
              RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.
              
              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 66, 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.

ChemVL Data

Dataset repository for the ChemVL chemistry vision-language model (code on GitHub).
This snapshot provides pretrained weights, downstream datasets (MoleculeNet / MoleculeACE as 2D structure images), finetuned checkpoints, descriptor metadata, and knowledge-memory caches needed to reproduce the paper and run the official tutorials.

Download

pip install huggingface_hub
export CHEMVL_DATA_ROOT=/path/to/chemvl-data

git clone https://github.com/yhuang1997/ChemVL.git
cd ChemVL
python tools/hf_download.py download
python tools/hf_download.py unpack

Set CHEMVL_DATA_ROOT to the directory where the snapshot was downloaded. Large directory trees are shipped as **.tar.zst archives** under archives/; hf_download.py unpack restores the layout expected by the training code.

Layout (after unpack)

CHEMVL_DATA_ROOT/
├── descriptor_info.pkl
├── archives/                              # download artifacts (optional to delete after unpack)
├── cache_for_knowledge/*.pkl
├── checkpoints/
│   ├── pretraining/RN50px224.ckpt         # image backbone
│   ├── pretraining/GIN.ckpt               # graph backbone
│   ├── external/                          # MolCLR, ImageMol baselines
│   └── finetuning/                        # from archives/checkpoints_finetuning.tar.zst
├── finetuning_datasets/                   # from archives/finetuning_datasets.tar.zst
│   ├── MPP/classification/                # MoleculeNet (6 cls tasks)
│   ├── MPP/regression/                    # MoleculeNet (4 reg tasks)
│   └── MoleculeACE/                       # 30 ChEMBL targets
└── pretraining_datasets/
    ├── 10M-106mds/                        # graph + image pretrain metadata (on Hub)
    └── images-10M@224px/                  # generate locally (see below)

Update literal paths in JSON configs (dataset.dataroot, model.resume) to match your CHEMVL_DATA_ROOT.

Image pretraining corpus (generate locally)

The ~10M PNG corpus for ChemVL-Image pretraining is not stored on the Hub (too large). The Hub snapshot includes pretraining_datasets/10M-106mds/ (mds.csv, train.txt, test.txt). Generate 224×224 PNGs with the ChemVL repo (requires rdkit-pypi==2022.9.5, same as the main README):

export CHEMVL_DATA_ROOT=/path/to/chemvl-data

python tools/datasets/render_pretrain_images.py render --split train --skip-existing
python tools/datasets/render_pretrain_images.py render --split test --skip-existing

Output layout (matches ordinalclip/configs/base_cfgs/data_cfg/datasets/mol-10M-106mds/local.yaml):

pretraining_datasets/images-10M@224px/train_data/{index}.png
pretraining_datasets/images-10M@224px/test_data/{index}.png

Rendering uses utils/pretrain_image_render.py (MolFromSmiles + RDKit MolsToGridImage at 224px), the same recipe as default downstream PNGs in ChemVL.

Category	What you get
Pretrained weights	ChemVL-Image (RN50), ChemVL-Graph (GIN)
Finetuned weights	Preset checkpoints under `checkpoints/finetuning/presets/` (config + weights)
Downstream data	Full MoleculeNet-10 and MoleculeACE-30 224×224 PNGs
Pretraining	`10M-106mds` metadata on Hub; image PNGs via `render_pretrain_images.py`
Other	`descriptor_info.pkl`; MoleculeNet `cache_for_knowledge/*.pkl` (no MoleculeACE caches)

Quick start (after download)

# MoleculeNet BBBP fine-tuning
python extensive_finetune.py \
  --config configs/tutorials/moleculenet_bbbp_classification_scaffold_prior_guided.json

# MoleculeACE fine-tuning
python moleculeace_finetune.py \
  --config configs/tutorials/moleculeace_chembl2047_ec50_molmcl_regression.json

See the GitHub README.md for environment setup (Python 3.9, PyTorch 1.13.1+cu117, RDKit). A GPU is required for fine-tuning and inference.

Downloads last month: 5

Total file size:

59.9 GB

ChemVL Data

Download

Layout (after unpack)

Image pretraining corpus (generate locally)

Contents

Quick start (after download)