DeePMD-kit v3: A Multiple-Backend Framework for Machine Learning Potentials

JZ Zeng and D Zhang and AY Peng and XY Zhang and SS He and Y Wang and XZJ Liu and HR Bi and YF Li and C Cai and CQ Zhang and YM Du and JX Zhu and PH Mo and ZT Huang and QY Zeng and SC Shi and XJ Qin and ZX Yu and CX Luo and Y Ding and YP Liu and RS Shi and ZY Wang and SL Bore and JH Chang and Z Deng and ZH Ding and SY Han and WR Jiang and GL Ke and ZQ Liu and DH Lu and K Muraoka and H Oliaei and AK Singh and HH Que and WH Xu and ZMC Xu and YB Zhuang and JY Dai and TJ Giese and WL Jia and B Xu and DM York and LF Zhang and H Wang, JOURNAL OF CHEMICAL THEORY AND COMPUTATION, 21, 4375-4385 (2025).

DOI: 10.1021/acs.jctc.5c00340

In recent years, machine learning potentials (MLPs) have become indispensable tools in physics, chemistry, and materials science, driving the development of software packages for molecular dynamics (MD) simulations and related applications. These packages, typically built on specific machine learning frameworks, such as TensorFlow, PyTorch, or JAX, face integration challenges when advanced applications demand communication across different frameworks. The previous TensorFlow-based implementation of the DeePMD-kit exemplified these limitations. In this work, we introduce DeePMD-kit version 3, a significant update featuring a multibackend framework that supports TensorFlow, PyTorch, JAX, and PaddlePaddle backends, and demonstrate the versatility of this architecture through the integration of other MLP packages and of differentiable molecular force fields. This architecture allows seamless back-end switching with minimal modifications, enabling users and developers to integrate DeePMD-kit with other packages using different machine learning frameworks. This innovation facilitates the development of more complex and interoperable workflows, paving the way for broader applications of MLPs in scientific research.

Return to Publications page