There is a newer version of the record available.

Published April 7, 2026 | Version v1
Journal article Open

Dataset and models supporting "NEP89: Universal neuroevolution potential for inorganic and organic materials across 89 elements"

  • 1. The Chinese University of Hong Kong
  • 2. Bohai university
  • 3. ROR icon Chalmers University of Technology
  • 4. Shenzhen Technology University
  • 5. Xinyu Universiy
  • 6. ROR icon University of Science and Technology Beijing
  • 7. ROR icon Bohai University
  • 8. Aalto University
  • 9. ROR icon Fuzhou University
  • 10. ROR icon Tel Aviv University
  • 11. ROR icon Zhejiang University
  • 12. ROR icon George Washington University

Description

Description

This repository contains the dataset, source code, demonstration files, and fine-tuning workflows for NEP89, a universal Neuroevolution Potential (NEP) foundation model covering 89 elements across the periodic table, as described in Liang et al., "NEP89: Universal neuroevolution potential for inorganic and organic materials across 89 elements" [arXiv:2504.21286].

Included Files

  • Dataset_for_NEP89.zip This archive contains the comprehensive dataset used to train the NEP89 model. It comprises carefully subsampled structures from diverse sources (e.g., OMAT24, MPtrj, SPICE, ANI-1xnr, UNEP-v1, solid-state electrolytes, water, protein, and CH systems), along with newly constructed CHONPS data. The dataset is organized into 11 distinct subsets, which can be used independently or combined into a unified dataset for machine-learned potential (MLP) training and development.

  • GPUMD-5.0-NEP89-Demos.zip This archive provides the essential software environment and practical examples for the NEP89 model, including:

    • Source code for NEP89 model deployment and performing Molecular Dynamics (MD) simulations.

    • Out-of-the-box demos for rapid testing of the NEP89 model.

    • Fine-tuning workflows for adapting the NEP89 model to specific chemical environments or high-accuracy requirements.

Quick Links & Resources

For the most up-to-date resources and detailed instructions, please refer to:

  • NEP89 Model: The latest NEP89 potential files are available in the GPUMD GitHub Repository.

  • Fine-tuning Tutorial: A step-by-step guide to fine-tuning NEP89 is available in the GPUMD-Tutorials.

  • Software Requirement: For optimal performance and compatibility, especially during fine-tuning, we recommend using the latest version of GPUMD (version 5.0).

Additional Information

Further technical details and implementation notes are provided in the README files within each archive. 

For repository- and implementation-related inquiries, please contact: 
liangting.zj@gmail.com, phychensd@gmail.com, brucenju@gmail.com

Files

GPUMD-5.0-NEP89-Demos.zip

Files (526.0 MB)

Name Size Download all
md5:d01c9c794718d53621a83c68c3cf194f
453.8 MB Preview Download
md5:1f37b08eb334d60f4bd2338a6607751d
72.2 MB Preview Download

Additional details

Related works

Is supplemented by
Publication: https://arxiv.org/abs/2504.21286 (URL)

Dates

Available
2026-04-07

Software

Repository URL
https://gpumd.org
Programming language
Cuda