DeePMD-kit#
Start from a pretrained Deep Potential model, fine-tune it for your system, and deploy it at simulation scale.
Important
A pretrained model can be your starting point, not just your end result. Download a built-in DPA checkpoint, fine-tune the full model, or use DPA-4 LoRA with PyTorch single-task training, then test, export, and deploy it through the same DeePMD-kit workflow.
DeePMD-kit turns quantum-mechanical reference data into fast, scalable interatomic potentials. Use it across molecular and materials science—from finite molecules and covalent systems to periodic solids and metals—and scale from laptop fine-tuning to distributed training and MPI-parallel molecular dynamics.
The DPA4 model family traces a Pareto frontier across Matbench Discovery CPS and saturated inference throughput.#
Choose your path#
Download a built-in DPA checkpoint or resolve its model name directly.
Adapt a full pretrained model, or use DPA-4 LoRA with PyTorch single-task training.
Build a new potential with single-task, multi-task, or distributed training.
Compare DeepPot-SE, DPA-1, DPA-2, DPA-3, DPA-4, and specialized physics models.
Move models into Python, native APIs, LAMMPS, i-PI, ASE, GROMACS, and the wider simulation ecosystem.
Why DeePMD-kit#
Start from built-in DPA checkpoints, fine-tune full models, use DPA-4 LoRA with PyTorch single-task training, and reuse learned representations with DPA-ADAPT.
Configure new architectures and physical targets, then train with single-task, multi-task, or distributed workflows.
Use efficient DeepPot-SE descriptors, attention-based DPA models, large atomic models, and SO(3)-equivariant DPA-4.
Learn energies, forces, virials, Hessians, spin, dipoles, polarizabilities, density of states, atomic populations, and custom properties.
Work with TensorFlow, PyTorch, JAX, or Paddle and use backend-aware model formats, conversion, and runtime plugins.
Run on CPUs and GPUs, distribute training, compress supported models, export compiled .pt2 artifacts, and drive MPI-parallel simulations.
Connect to simulation engines, workflow tools, native applications, and external GNN models through documented interfaces and plugins.
Tip
On supported descriptors and workloads, model compression can deliver more than 10× inference speedup and reduce memory usage by as much as 20×. Actual gains depend on the model, system, and hardware.
Two starting points, one path to dynamics#
Select a built-in DPA model or a compatible checkpoint.
Convert target structures and labels into DeePMD data.
Adapt the full model, or use DPA-4 LoRA with PyTorch single-task training.
Build a new model from scratch when adaptation is not the right fit.
Test accuracy, inspect deviation, freeze, and compress.
Run inference directly or deploy into molecular dynamics.
Fine-tuning and from-scratch training converge on the same validation, export, and deployment toolchain. DPA-4 LoRA is currently limited to PyTorch single-task fine-tuning.
Choose a model family#
Family | A strong starting point when you need |
|---|---|
An efficient, established baseline with broad backend and deployment support. | |
Attention-based local representations and type embedding. | |
Multi-task pretraining, shared representations, and smooth conservative potentials. | |
Message passing over line-graph representations and broad chemical coverage. | |
SO(3)-equivariant learning, DPA-4 LoRA for PyTorch single-task fine-tuning, optional ZBL bridging, spin support, and compiled |
DPA4 provides a family of accuracy–throughput trade-offs for different deployment budgets.#
More than conventional force fields#
Train spin-aware potentials with atomic and magnetic force targets.
Combine learned local interactions with DPLR electrostatics, DPRc range correction, pair tables, or analytical ZBL bridging.
Predict electronic or structural properties and export learned representations for analysis or downstream models.
New and noteworthy#
Resolve built-in model names directly or download checkpoints to a local cache.
Equivariant message passing, PyTorch single-task LoRA, ZBL, spin, compiled inference, and LAMMPS deployment.
Adapt pretrained DPA representations to downstream atomistic property tasks.
Give AI agents reproducible guidance for training, fine-tuning, inference, and LAMMPS workflows.
Documentation map#
Getting Started: Installation and first model.
User Guide: Backends · Installation · Data · Models · Training · Freeze · Test · Inference · DPA-ADAPT · CLI · Integrations · Agent Skills · NVNMD · Environment · Troubleshooting.
Tutorials and Publications: Tutorials · Publications.
Developer Guide: CMake · TensorFlow models · PyTorch models · Type embedding · Coding conventions · CI/CD · Python API · Custom operators · C++ API · C API · Core API.
Project Details: License · Authors and credits · Logo.
Citation#
If you use DeePMD-kit in published work, cite the general software publication that matches the version used:
Han Wang, Linfeng Zhang, Jiequn Han, and Weinan E. “DeePMD-kit: A deep learning package for many-body potential energy representation and molecular dynamics.” Computer Physics Communications 228 (2018): 178–184. DOI: 10.1016/j.cpc.2018.03.016.
Jinzhe Zeng, Duo Zhang, Denghui Lu, Pinghui Mo, Zeyu Li, Yixiao Chen, Marián Rynik, Li’ang Huang, Ziyao Li, Shaochen Shi, Yingze Wang, Haotian Ye, Ping Tuo, Jiabin Yang, Ye Ding, Yifan Li, Davide Tisi, Qiyu Zeng, Han Bao, Yu Xia, Jiameng Huang, Koki Muraoka, Yibo Wang, Junhan Chang, Fengbo Yuan, Sigbjørn Løland Bore, Chun Cai, Yinnian Lin, Bo Wang, Jiayan Xu, Jia-Xin Zhu, Chenxing Luo, Yuzhi Zhang, Rhys E. A. Goodall, Wenshuo Liang, Anurag Kumar Singh, Sikai Yao, Jingchao Zhang, Renata Wentzcovitch, Jiequn Han, Jie Liu, Weile Jia, Darrin M. York, Weinan E, Roberto Car, Linfeng Zhang, and Han Wang. “DeePMD-kit v2: A software package for Deep Potential models.” The Journal of Chemical Physics 159 (2023): 054801. DOI: 10.1063/5.0155600.
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, and Han Wang. “DeePMD-kit v3: A Multiple-Backend Framework for Machine Learning Potentials.” Journal of Chemical Theory and Computation 21 (2025): 4375–4385. DOI: 10.1021/acs.jctc.5c00340.
Follow the citation guide for the method-specific publications required by the models and features used in your work.
Note
DeePMD-kit is licensed under the GNU LGPL-3.0-or-later.