Quickstart

This guide gets you from installation to first results.

1. Install

  • Python 3.10–3.12

  • Recommended: create a fresh environment

conda create -n nepkit python=3.10
conda activate nepkit
pip install NepTrainKit

GPU build note (Linux/WSL2): pip install auto‑detects CUDA. If CUDA is not detected, export one of CUDA_HOME or CUDA_PATH and ensure lib64 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) explicitly target your GPU compute capability (SM) when compiling nep_gpu
# 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"
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

Windows portable: download NepTrainKit.win32.zip from Releases and run the executable.

2. Launch

nepkit
# or
NepTrainKit

3. NEP Dataset Display

  • Import data via the top‑left Open button or drag‑and‑drop.

  • Supported imports:

    • train.xyz + corresponding *.out files

    • nep.txt (optional; uses NEP89 if absent) + train.xyz

    • DeepMD directory (auto‑detected)

  • Interact with plots, search by tag/formula/elements, select, delete, and export:

    • Export menu → “Export Selected Structures” for chosen frames

    • Save button exports export_remove_model.xyz and export_good_model.xyz

4. Make Dataset

  • Drag structures (XYZ/POSCAR/CIF) into the window or use Open.

  • Build a pipeline with cards; use groups to branch/merge; add FPS filter if needed.

  • Export to make_dataset.xyz when done.

  • Save/Load card configurations as JSON to reuse pipelines.

5. Data Management

  • Organize datasets into Projects and Models (versions), with notes and tags.

  • Right‑click for New/Modify/Delete, Open Folder, and Tag management.

  • Press Ctrl+F for advanced search.

6. Settings

  • Choose plotting force mode (Raw vs Norm) and canvas engine (PyQtGraph vs Vispy).

  • NEP Backend: select CPU/GPU/Auto for NEP calculations; Auto tries GPU first and falls back to CPU

  • GPU Batch Size: adjust the number of frames per GPU slice to balance speed and memory

  • Enable Auto loading, adjust covalent radius threshold, sorting, and menu grouping.

  • Check app updates and NEP89 model, open help and feedback.

Note:

  • GPU backend requires a compatible NVIDIA driver and CUDA 12.4 runtime. If you see “CUDA driver version is insufficient for CUDA runtime version”, switch NEP Backend to CPU in Settings.

7. Tips

  • Use Vispy for large scenes if your GPU supports OpenGL.

  • Switch search mode to formula to match by composition rather than tags.

  • Switch search mode to elements to filter by element set, e.g. Fe,O (only Fe/O present) or +Fe (must contain Fe).

  • Use the structure toolbar to export descriptors or mark non‑physical bonds.

Cite NepTrainKit

If you publish results that rely on NepTrainKit, cite the following paper and acknowledge upstream NEP projects where relevant:

@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},
}