nayoung10/mofflow2-ckpt

MOFFlow-2 Pretrained Checkpoints

Pretrained checkpoints and configs for MOFFlow-2 tasks:

• Structure prediction (csp)
• Generation (gen)

Project repo: https://github.com/nayoung10/MOFFlow-2

Files

• csp/config.yaml
• csp/sp_module_ckpt.tar.gz
• gen/sp_module/config.yaml
• gen/sp_module/sp_module_ckpt.tar.gz
• gen/seq_module_unconditional/config.yaml
• gen/seq_module_unconditional/seq_module_unconditional_ckpt.tar.gz
• gen/seq_module_conditional/config.yaml
• gen/seq_module_conditional/seq_module_conditional_ckpt.tar.gz
• SHA256SUMS

Quick Start (download + verify + extract)

# Download all files from this model repo
hf download lkny123/mofflow2-ckpt --repo-type model --local-dir .

# Verify integrity
sha256sum -c SHA256SUMS

Extract checkpoint archives in place:

tar -xzf csp/sp_module_ckpt.tar.gz -C csp
tar -xzf gen/sp_module/sp_module_ckpt.tar.gz -C gen/sp_module
tar -xzf gen/seq_module_unconditional/seq_module_unconditional_ckpt.tar.gz -C gen/seq_module_unconditional
tar -xzf gen/seq_module_conditional/seq_module_conditional_ckpt.tar.gz -C gen/seq_module_conditional

How to use with MOFFlow-2

Important: config.yaml must be in the same directory as the checkpoint file used in inference.ckpt_path.

Example:

python -m experiments.predict \
  inference.task=csp \
  inference.ckpt_path=<path_to_ckpt_file_inside_csp_dir>

For sequence generation:

python -m experiments.predict_seq \
  inference.ckpt_path=<path_to_ckpt_file_inside_gen/seq_module_unconditional_or_conditional>

Notes

• Checkpoint files are PyTorch Lightning .ckpt.
• Archive members may include epoch/step in filename.
• Use the extracted checkpoint path directly in inference.ckpt_path.