Meta FAIR Releases Open Materials 2024 Dataset with 110M+ DFT Calculations
Meta FAIR has unveiled the Open Materials 2024 (OMat24) dataset, which is a comprehensive open resource featuring more than 110 million density functional theory (DFT) calculations that emphasize structural and compositional variety. Alongside this release, a collection of pre-trained EquiformerV2 models has been introduced, demonstrating top-tier performance on the Matbench Discovery leaderboard. The objective of this dataset is to expedite AI-enhanced materials discovery and design by offering accessible training data and open models, tackling a significant obstacle in the domain. This effort has the potential to influence areas ranging from climate change solutions to advanced computing hardware.
Key facts
- Meta FAIR released Open Materials 2024 (OMat24) dataset.
- OMat24 contains over 110 million DFT calculations.
- Dataset focuses on structural and compositional diversity.
- EquiformerV2 models achieve state-of-the-art on Matbench Discovery.
- Release includes pre-trained models.
- Aims to accelerate AI-driven materials discovery.
- Addresses lack of public training data and open models.
- Potential applications in climate change and computing hardware.
Entities
Institutions
- Meta FAIR