Description
This is an OMatG (Open Materials Generation) model for de-novo generation (DNG) of inorganic crystals trained on the MP-20 dataset.
The MP-20 dataset consists of structures from the Materials Project database containing 20 or fewer atoms. This dataset is also included in OMatG.
The subdirectories in this repository contain various model hyperparameters and training checkpoints for a variety of MP-20-DNG models.
Uses
The checkpoints and model hyperparameters can be used for de-novo generation of crystalline structures with OMatG, as described in the this README.md file
Recommendations
The Linear-SDE-Gamma checkpoints currently provide the best results with respect to structure stability, novelty, and uniqueness.
Citation
Please cite our paper on OpenReview if using OMatG.
Links
OMatG on GitHub: See this repository for OMatG installation, training and usage instructions.
KIM Initiative: Knowledgebase of Interatomic Models. Tools and resources for researchers in materials science and chemistry.
Fermat-ML on GitHub: Foundational Representation of Materials. Machine learning foundation model for materials and chemistry discovery.