The rapid advancements in artificial intelligence (AI) are catalyzing transformative changes in atomic modeling, simulation, and design. AI-driven potential energy models have demonstrated the capability to conduct large-scale, long-duration simulations with the accuracy of ab initio electronic structure methods. However, the model generation process remains a bottleneck for large-scale applications. We propose a shift towards a model-centric ecosystem, wherein a large atomic model (LAM), pre-trained across multiple disciplines, can be efficiently fine-tuned and distilled for various downstream tasks, thereby establishing a new framework for molecular modeling. In this study, we introduce the DPA-2 architecture as a prototype for LAMs. Pre-trained on a diverse array of chemical and materials systems using a multi-task approach, DPA-2 demonstrates superior generalization capabilities across multiple downstream tasks compared to the traditional single-task pre-training and fine-tuning methodologies. Our approach sets the stage for the development and broad application of LAMs in molecular and materials simulation research.
@article{arxiv.2312.15492,
title = {DPA-2: a large atomic model as a multi-task learner},
author = {Duo Zhang and Xinzijian Liu and Xiangyu Zhang and Chengqian Zhang and Chun Cai and Hangrui Bi and Yiming Du and Xuejian Qin and Anyang Peng and Jiameng Huang and Bowen Li and Yifan Shan and Jinzhe Zeng and Yuzhi Zhang and Siyuan Liu and Yifan Li and Junhan Chang and Xinyan Wang and Shuo Zhou and Jianchuan Liu and Xiaoshan Luo and Zhenyu Wang and Wanrun Jiang and Jing Wu and Yudi Yang and Jiyuan Yang and Manyi Yang and Fu-Qiang Gong and Linshuang Zhang and Mengchao Shi and Fu-Zhi Dai and Darrin M. York and Shi Liu and Tong Zhu and Zhicheng Zhong and Jian Lv and Jun Cheng and Weile Jia and Mohan Chen and Guolin Ke and Weinan E and Linfeng Zhang and Han Wang},
journal= {arXiv preprint arXiv:2312.15492},
year = {2025}
}