Open MatSciML Toolkit

⊕ Open source, AI-driven materials discovery framework and benchmark

In order to meet the rapidly growing societal demands for better, sustainable materials, we need to apply AI to accelerate the chemical discovery process.

The Open MatSciML Toolkit is an open source framework developed at Intel Labs that has several overlapping goals:

1. Lower the barrier to entry for experimenting with AI-centric workflows in materials science.

2. Implement state-of-the-art models for evaluation and use.

3. Improve reproducibility, performance portability, and reduce technical debt for bleeding edge AI in materials science and chemistry.

Development of the framework is ongoing, and work that utilizes it has been featured in a few venues including the AI4Mat at NeurIPS 2023, and the AI4Science workshop at Super Computing 2023. The framework is research driven; we have experimented with and implemented features in support of developing foundation models (e.g. pretraining, multitask, multitdata functionality), and AI-based interatomic potentials that are for molecular dynamics simulations.