ChemBFN: A Bayesian Flow Network framework for Chemistry

Welcome to the model weights of ChemBFN model (arXiv version: arxiv.org/abs/2407.20294 & arxiv.org/abs/2412.11439). You can find the code on our GitHub page.

Pre-trained models

• zinc15_40m.pt: a 54M ChemBFN model trained on 40M molecules (SMILES) from ZINC15 database.
• zinc15_190m.pt: a 54M ChemBFN model trained on 190M molecules (SMILES) from ZINC15 database.
• zinc15_40m_selfies.pt: a 54M ChemBFN model trained on 40M molecules (SELFIES) from ZINC15 database.
• zinc15_190m_large.pt: a 217M ChemBFN model trained on 190M molecules (SMILES) from ZINC15 database.
• zinc15_40m_safe_large.pt: a 217M ChemBFN model trained on 40M molecules (SAFE) from ZINC15 database.

Configuration file

• config.json stores the configuration settings of pretrained base models.

Vocabulary file

• zinc15_selfies_vocab.txt stores the SELFIES vocabulary.

Model Card Author

Nianze TAO