ShEPhERD v0.2.4

What's Changed

v0.2.4: Model loading and repository optimization

• Added automatic model downloading from HuggingFace Hub with load_model(), get_model_info(), and clear_model_cache() functions
• Removed model weights from git history to reduce repository size - users should re-clone the repository
• Added ability for interrupting inference with improved UI to Streamlit app

v0.2.3: Add Streamlit app for demonstrations

• Added an easy-to-use app for demonstration purposes

v0.2.2: Refresh inference code

• Refactored ShEPhERD inference code to be more modular (backwards compatible).
  • The original inference code can still be imported: from shepherd.inference import inference_sample
  • New inference functions can be imported with:
    • from shepherd.inference import generate
    • from shepherd.inference import generate_from_intermediate_time
• Inference now supports atom and bond inpainting
  • generate is updated to allow atom inpainting from t=T
  • generate_from_intermediate_time is specialized to allow atom inpainting from an intermediate time (T ≤ t < 0)
• Inference can store full diffusion trajectories by setting return_trajectories=True during sampling.
• shepherd.extract
  • Added remove_side_groups_with_geometry
  • Added remove_overlaps to quickly filter sampled molecules that use atom-inpainting.

Full Changelog: v0.2.0...v0.2.4

ShEPhERD v0.2.0: Refactor

Refactoring to support PyTorch >= v2.5.1

Merged from #11

• Refactored ShEPhERD as a package
  • Updated import statements: throughout repo to import directly from shepherd assuming local install.
  • Fix depreciation warnings:
    • torch.load() -> torch.load(weights_only=True)
    • @torch.cuda.amp.autocast(enabled=False) -> @torch.amp.autocast('cuda', enabled=False)
  • Training scripts
    • Updated src/shepherd/datasets.py for higher versions of PyG. Required changes to the batching functionality for edges (still backwards compatible).
    • Slight changes to training/train.py for upgraded versions of PyTorch Lightning.
• Model checkpoints have been UPDATED for PyTorch Lightning v2.5.1
  • The original checkpoints for PyTorch Lightning v1.2 can be found in previous commits (c3d5ec0 or before), the original publication Release, or at the Dropbox data link: https://www.dropbox.com/scl/fo/rgn33g9kwthnjt27bsc3m/ADGt-CplyEXSU7u5MKc0aTo?rlkey=fhi74vkktpoj1irl84ehnw95h&e=1&st=wn46d6o2&dl=0
• Created a basic unconditional generation test script
• Updated the environment and relevant files to be compatible with PyTorch >= v2.5.1
• Bug fix for shepherd.datasets.HeteroDataset.__getitem__ where x3 point extraction should use get_x2_data

Additional notes

Thank you to Matthew Cox for his contributions in the updated code.

Full Changelog: Publication...v0.2.0-alpha

ShEPhERD v0.1.0: Publication code

Publication code

This release contains the code and weights used in the original publication (https://arxiv.org/abs/2411.04130).

In particular, ShEPhERD was trained and tested with the following package versions:

python==3.8.13
pytorch==1.12.1
pyg==2.2.0
pytorch-lightning==1.6.3
cudatoolkit==11.3.1