UMA: A Family of Universal Models for Atoms

2025-06-30

AI Summary

High-throughput atomic-scale simulations are urgently needed in chemistry and materials science, yet existing models struggle to jointly optimize prediction accuracy, inference speed, and cross-domain generalization. Method: We propose a universal atomic modeling framework based on a hybrid Mixture-of-Experts (MoE) architecture, enabling scalable model capacity (up to 1.4 billion parameters) while maintaining low activated parameter count per structure (~50 million). The model is trained on diverse 3D structural data spanning molecules, bulk materials, and heterogeneous catalysts, and an empirically validated scaling law is identified to support general-purpose, cross-task modeling. Results: Without task-specific fine-tuning, this single unified model matches or surpasses specialized models across multiple downstream tasks—including energy, force, and electronic property prediction—demonstrating strong zero-shot transferability. It advances an efficient, scalable, and plug-and-play paradigm for atomistic computation.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Cognitive Modeling & Cognitive Systems: (Computational) Cognitive Architectures
• Computer Vision: Large Vision Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

The ability to quickly and accurately compute properties from atomic simulations is critical for advancing a large number of applications in chemistry and materials science including drug discovery, energy storage, and semiconductor manufacturing. To address this need, Meta FAIR presents a family of Universal Models for Atoms (UMA), designed to push the frontier of speed, accuracy, and generalization. UMA models are trained on half a billion unique 3D atomic structures (the largest training runs to date) by compiling data across multiple chemical domains, e.g. molecules, materials, and catalysts. We develop empirical scaling laws to help understand how to increase model capacity alongside dataset size to achieve the best accuracy. The UMA small and medium models utilize a novel architectural design we refer to as mixture of linear experts that enables increasing model capacity without sacrificing speed. For example, UMA-medium has 1.4B parameters but only ~50M active parameters per atomic structure. We evaluate UMA models on a diverse set of applications across multiple domains and find that, remarkably, a single model without any fine-tuning can perform similarly or better than specialized models. We are releasing the UMA code, weights, and associated data to accelerate computational workflows and enable the community to continue to build increasingly capable AI models.

Problem

• Develop fast, accurate universal atomic property prediction models
• Scale model capacity with dataset size for optimal accuracy
• Enable cross-domain applications without specialized model fine-tuning

Innovation

• Universal Models for Atoms (UMA) family
• Mixture of linear experts architecture
• Empirical scaling laws for model capacity