What Can DP Do too? | DeePMD - kit v3 Paper Published: Multi - backend Framework as the Innovation
Recently, the Journal of Chemical Theory and Computation published a research work titled DeePMD - kit v3: A Multiple - Backend Framework for Machine Learning Potentials [1]. This work focuses on a core innovation in the DeePMD - kit v3 version - the multi - backend framework. The latest version has integrated four deep learning backends: TensorFlow, PyTorch, JAX, and PaddlePaddle.
The paper elaborates on the design concept and implementation details of the multi - backend framework. Different from other software that independently implements different backends in separate software packages, DeePMD - kit enables multiple backends to share a unified module through a clever structural design, significantly reducing development and maintenance costs. In addition, the paper compares the characteristics of different backends and introduces other important design details in the v3 version.
The paper also highlights two major advantages of the multi - backend framework. In terms of functionality, taking the DeePMD - GNN plugin [2] and the DMFF plugin as examples, the paper shows how the multi - backend architecture better supports the expansion of new functions dependent on specific backends. In terms of performance, the research team conducted comprehensive dynamic simulation performance tests on backends such as TensorFlow, PyTorch, and JAX. The results show that no single backend has an absolute advantage in all scenarios, allowing users to freely choose the most suitable backend according to specific needs to achieve optimal performance.
A total of 47 researchers from 41 institutions in 7 countries participated in this paper. Among them, Jinzhe Zeng, a tenure - track associate professor at the School of Artificial Intelligence and Data Science, University of Science and Technology of China, is the first author and co - corresponding author of the paper. Linfeng Zhang, Dean of the Beijing Academy of Scientific Intelligence and Founder and Chief Scientist of DeepSeek, and Han Wang, a researcher at the Beijing Institute of Applied Physics and Computational Mathematics, also serve as co - corresponding authors. The School of Artificial Intelligence and Data Science, University of Science and Technology of China, is the first unit of the paper.
Paper address: https://doi.org/10.1021/acs.jctc.5c00340
Usage method
The usage method of the multi - backend framework was demonstrated in last year's tutorial. Click https://mp.weixin.qq.com/s/nLlVMXoOvh59PX1Jfm1eJA to go.
Talent recruitment
Jinzhe Zeng, the first author and co - corresponding author of this paper, is a tenure - track associate professor and doctoral supervisor at the School of Artificial Intelligence and Data Science, University of Science and Technology of China. He is recruiting outstanding undergraduates to pursue a doctoral degree (major: 140500 Intelligent Science and Technology). Contact information: jinzhe.zeng@ustc.edu.cn
References
[1] Jinzhe Zeng, Duo Zhang, Anyang Peng, Xiangyu Zhang, Sensen He, Yan Wang, Xinzijian Liu, Hangrui Bi, Yifan Li, Chun Cai, Chengqian Zhang, Yiming Du, Jia - Xin Zhu, Pinghui Mo, Zhengtao Huang, Qiyu Zeng, Shaochen Shi, Xuejian Qin, Zhaoxi Yu, Chenxing Luo, Ye Ding, Yun - Pei Liu, Ruosong Shi, Zhenyu Wang, Sigbjørn Løland Bore, Junhan Chang, Zhe Deng, Zhaohan Ding, Siyuan Han, Wanrun Jiang, Guolin Ke, Zhaoqing Liu, Denghui Lu, Koki Muraoka, Hananeh Oliaei, Anurag Kumar Singh, Haohui Que, Weihong Xu, Zhangmancang Xu, Yong - Bin Zhuang, Jiayu Dai, Timothy J. Giese, Weile Jia, Ben Xu, Darrin M. York, Linfeng Zhang, Han Wang, DeePMD - kit v3: A Multiple - Backend Framework for Machine Learning Potentials, Journal of Chemical Theory and Computation, 2025.
[2] Jinzhe Zeng, Timothy J. Giese, Duo Zhang, Han Wang, Darrin M. York, DeePMD - GNN: A DeePMD - kit Plugin for External Graph Neural Network Potentials, Journal of Chemical Information and Modeling, 2025, 65, 7, 3154 - 3160.