torchmd/torchmd-net

Training neural network potentials

What it solves

TorchMD-NET provides high-performance neural network potentials (NNPs) used to model the energy and forces of molecular systems. It allows researchers to train these potentials on molecular datasets and integrate them into GPU-accelerated molecular dynamics (MD) simulations, replacing or augmenting traditional physics-based force fields.

How it works

The project exposes various neural network architectures as PyTorch modules. It supports state-of-the-art architectures like TensorNet and TensorNet2, and previously supported Equivariant Transformers and Graph Neural Networks. Users can train models using configuration YAML files or command-line arguments, with support for multi-node, multi-GPU training via PyTorch Lightning.

Who it’s for

It is designed for computational chemists, physicists, and machine learning researchers focusing on molecular dynamics and the development of neural network potentials.

Highlights

  • Integration: Works with GPU-accelerated MD codes such as ACEMD, OpenMM, and TorchMD.
  • Flexible Training: Supports multi-node training and custom dataset implementation via NumPy or torch-geometric.
  • Pretrained Models: Provides access to AceFF models via Hugging Face.
  • Extensibility: Includes a developer guide for implementing new neural network architectures within the framework.

Related

  • Project
    brucefan1983/GPUMD

    A highly efficient, GPU-accelerated molecular dynamics package that enables the training and using of neuroevolution machine-learned potentials for large-scale atomistic simulations.

  • Project
    BindsNET/bindsnet

    A PyTorch-based library for simulating spiking neural networks (SNNs) to develop biologically inspired machine learning and reinforcement learning algorithms.

  • Project
    TorchSim/torch-sim

    An open-source atomistic simulation engine built in PyTorch that accelerates Machine Learning Interatomic Potentials (MLIPs) through automatic batching and GPU acceleration.

  • Project
    pytorch/pytorch

    A Python-based machine learning framework that provides GPU-accelerated tensor computation and a dynamic autograd system for building deep neural networks.

  • Project
    aiqm/torchani

    TorchANI 2.0 is a PyTorch library for training and using ANI‑style neural‑network interatomic potentials, providing GPU‑accelerated energy/force predictions and tools for molecular dynamics and ML/MM simulations.