NepTrainKit 2.8.1

NepTrainKit is a Python package for visualizing and manipulating training datasets for NEP.

---

NepTrainKit

NepTrainKit is a toolkit focused on the operation and visualization of neuroevolution potential (NEP) training datasets. It is mainly used to simplify and optimize the NEP model training process, providing an intuitive graphical interface and analysis tools to help users adjust train dataset.

---

Community Support

• Join the community chat: https://qm.qq.com/q/wPDQYHMhyg
• Report issues or contribute via GitHub Issues

---

Installation

It is strongly recommended to use pip for installation.

Method 1: Install via pip

If you are using Python 3.10 or a later version, you can install NepTrainKit using an environment manager like conda:

1. Create a new environment:

   conda create -n nepkit python=3.10
   

2. Activate the environment:

   conda activate nepkit
   

3. For CentOS users, install PySide6 (required for GUI functionality):

   conda install -c conda-forge pyside6
   

• Install directly using the pip install command:

  pip install NepTrainKit
  

  GPU build note (Linux/WSL2): The build auto‑detects CUDA. If a compatible CUDA toolkit is present, the NEP backend is compiled with GPU acceleration; otherwise, it falls back to a CPU‑only build. If CUDA is not detected automatically, export one of CUDA_HOME or CUDA_PATH and ensure the lib64 directory is on your loader path before running pip install:

  # choose your installed CUDA version/path
  export CUDA_HOME=/usr/local/cuda-12.4
  export PATH="$CUDA_HOME/bin:$PATH"
  export LD_LIBRARY_PATH="$CUDA_HOME/lib64:${LD_LIBRARY_PATH}"
  
  # (optional) if you need to explicitly target your GPU compute capability (SM),
  # set NEP_GPU_GENCODE before pip install, e.g. for Turing (7.5):
  export NEP_GPU_GENCODE="arch=compute_75,code=sm_75"
  # or multiple targets:
  # export NEP_GPU_GENCODE="-gencode arch=compute_75,code=sm_75 -gencode arch=compute_86,code=sm_86"
  # now install
  pip install NepTrainKit
  

  GPU build note (Windows PowerShell): Set CUDA_PATH (or CUDA_HOME) and add bin to Path before pip install:

  $env:CUDA_PATH = "C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4"
  $env:Path = "$env:CUDA_PATH\\bin;" + $env:Path
  pip install NepTrainKit
  

  After installation, you can call the program using either NepTrainKit or nepkit.

• For the latest version (from GitHub):

  pip install git+https://github.com/aboys-cb/NepTrainKit.git
  

---

Method 2: Windows Executable

A standalone executable is available for Windows users.

• Visit the Releases page
• Download NepTrainKit.win32.zip

Note: Only supported on Windows platforms.

GPU Acceleration (optional)

• NepTrainKit includes an optional GPU‑accelerated NEP backend.
• Requirements: NVIDIA GPU/driver compatible with CUDA 12.4 runtime.
• Selection: In the app, go to Settings → NEP Backend and choose Auto/CPU/GPU.
  • Auto tries GPU first and falls back to CPU if unavailable.
  • Adjust GPU Batch Size to balance speed and memory.
  • If you see “CUDA driver version is insufficient for CUDA runtime version”, switch to CPU.

---

Documentation

For detailed usage documentation and examples, please refer to the official documentation:
https://neptrainkit.readthedocs.io/en/latest/index.html

• What's new: see docs/source/changelog.md or the Documentation "Changelog" page.

---

Citation

If you use NepTrainKit in academic work, please cite the following publication and acknowledge the upstream projects where appropriate:

@article{CHEN2025109859,
title = {NepTrain and NepTrainKit: Automated active learning and visualization toolkit for neuroevolution potentials},
journal = {Computer Physics Communications},
volume = {317},
pages = {109859},
year = {2025},
issn = {0010-4655},
doi = {https://doi.org/10.1016/j.cpc.2025.109859},
url = {https://www.sciencedirect.com/science/article/pii/S0010465525003613},
author = {Chengbing Chen and Yutong Li and Rui Zhao and Zhoulin Liu and Zheyong Fan and Gang Tang and Zhiyong Wang},
}

Licensing and Attribution

• License: This repository is licensed under the GNU General Public License v3.0 (or, at your option, any later version). See LICENSE at the repository root.
• Third‑party code: NepTrainKit incorporates source files and adapted logic from:
  • NEP_CPU (by Zheyong Fan, Junjie Wang, Eric Lindgren, and contributors): https://github.com/brucefan1983/NEP_CPU (GPL‑3.0‑or‑later)
  • GPUMD (by Zheyong Fan and the GPUMD development team): https://github.com/brucefan1983/GPUMD (GPL‑3.0‑or‑later)
• Directory‑level notes: See src/nep_cpu/README.md and src/nep_gpu/README.md for file‑level provenance, what was modified or added, and links to the upstream projects. A consolidated overview is also available in THIRD_PARTY_NOTICES.md.
• Redistribution: Any modifications and redistributions must remain under the GPL and preserve copyright and license notices, per the GPL requirements.

For academic use, cite NepTrainKit as shown above and acknowledge NEP_CPU and/or GPUMD as appropriate.