Position-based Equivariant Graph Neural Network (pos-egnn)

This repository contains PyTorch model for loading and performing inference using the pos-egnn, a foundation model for Chemistry and Materials.

GitHub: https://github.com/IBM/materials/tree/main/models/pos_egnn

HuggingFace: https://huggingface.co/ibm-research/materials.pos-egnn

Introduction

We present pos-egnn, a Position-based Equivariant Graph Neural Network foundation model for Chemistry and Materials. The model was pre-trained on 1.4M samples (i.e., 90%) from the Materials Project Trajectory (MPtrj) dataset to predict energies, forces and stress. pos-egnn can be used as a machine-learning potential, as a feature extractor, or can be fine-tuned for specific downstream tasks.

Besides the model weigths pos-egnn.v1-6M.pt (download from HuggingFace), we also provide an example.ipynb notebook (download from GitHub), which demonstrates how to perform inference, feature extraction and molecular dynamics simulation with the model.

For more information, please reach out to rneumann@br.ibm.com and/or flaviu.cipcigan@ibm.com