AI2Pot

Copyright © 2025 Hanyu Liu.

AI2Pot is distributed under the GNU General Public License v3.0.

What is AI2Pot?

AI2Pot (Ab Initio and Artificial Intelligence POTential) is a machine-learning interatomic-potential framework for materials modeling. It integrates multiple potential models, including the Moment Tensor Potential (MTP) and Neuroevolution Potential (NEP), within a unified PyTorch-based training and simulation workflow.

AI2Pot provides:

• Unified implementations of MTP and NEP
• High-performance C++ and CUDA operators
• PyTorch-based training, inference, and evaluation pipelines
• Interoperability with pymatgen, ASE, and LAMMPS
• A companion CLI tool, AI2Pot-cli, for data preprocessing, model training, and post-processing. (Repository: https://github.com/lhycms/AI2Pot-cli)

Installation

1. Configure CUDA

For a CUDA-enabled installation, set CUDA_HOME to the CUDA toolkit compatible with the installed PyTorch version:

$ export CUDA_HOME=/path/to/cuda
$ export PATH="${CUDA_HOME}/bin:${PATH}"
$ export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"

For example:

$ export CUDA_HOME=/usr/local/cuda-12.4

Check the CUDA compiler:

$ nvcc --version

2. Install PyTorch

Choose either the CPU or CUDA version.

Python environment

$ conda create -n ai2pot python=3.11.13

CPU

$ python -m pip install torch==2.5.0 \
    --index-url https://download.pytorch.org/whl/cpu

CUDA 12.4

$ python -m pip install torch==2.5.0 \
    --index-url https://download.pytorch.org/whl/cu124

Verify the PyTorch installation:

$ python -c "import torch; print(torch.__version__); print(torch.version.cuda)"

3. Install AI2Pot

AI2Pot currently needs to compile against the NumPy and PyTorch packages installed in the active Python environment. Therefore, build isolation must be disabled.

Install the build backend first:

$ python -m pip install -U pip setuptools wheel
$ python -m pip install \
    scikit-build-core==0.12.2 \
    cmake==4.3.2 \
    pybind11==2.11.1
$ python -m pip install -r requirements-lock.txt

Then install AI2Pot:

$ CMAKE_BUILD_PARALLEL_LEVEL=16 python -m pip install -v --no-build-isolation --no-deps .

CMAKE_BUILD_PARALLEL_LEVEL controls the number of parallel compilation jobs. Reduce this value if compilation consumes too much memory.

To use specific C, C++, and CUDA host compilers:

$ CMAKE_BUILD_PARALLEL_LEVEL=16 \
  CC=/path/to/gcc \
  CXX=/path/to/g++ \
  CUDAHOSTCXX=/path/to/g++ \
  python -m pip install -v --no-build-isolation --no-deps .

Developer installation and maintainer notes

Set the source-tree path

$ export AI2POT_PATH=/path/to/AI2Pot

Configure and build the native libraries

$ cd "${AI2POT_PATH}/source"
$ mkdir -p build
$ cd build

Configure the project:

$ cmake \
    -DBUILD_TEST=ON \
    -DTORCH_OP=ON \
    -DUSE_CUDA=ON \
    -DUSE_OPENMP=ON \
    -DMAX_MTP_LEVEL=20 \
    ..

Compile and install:

$ cmake --build . --parallel 16
$ cmake --install .

Equivalent Make commands:

$ make -j16
$ make install

Install the Python package in editable mode:

$ cd "${AI2POT_PATH}"
$ CMAKE_BUILD_PARALLEL_LEVEL=16 \
  CC=/data/app/gcc/11.3.0/bin/gcc \
  CXX=/data/app/gcc/11.3.0/bin/g++ \
  CUDAHOSTCXX=/data/app/gcc/11.3.0/bin/g++ \
  python -m pip install --no-build-isolation --no-deps -v -e .

Optional CMake arguments

-DCMAKE_C_COMPILER=/path/to/gcc
-DCMAKE_CXX_COMPILER=/path/to/g++
-DCMAKE_CUDA_HOST_COMPILER=/path/to/g++
-DCMAKE_INSTALL_PREFIX=$HOME/.ai2pot

Current development-server configuration

C compiler:         /data/app/gcc/11.3.0/bin/gcc
C++ compiler:       /data/app/gcc/11.3.0/bin/g++
CUDA host compiler: /data/app/gcc/11.3.0/bin/g++

Equivalent installation command:

$ CMAKE_BUILD_PARALLEL_LEVEL=16 \
  CC=/data/app/gcc/11.3.0/bin/gcc \
  CXX=/data/app/gcc/11.3.0/bin/g++ \
  CUDAHOSTCXX=/data/app/gcc/11.3.0/bin/g++ \
  python -m pip install -v --no-build-isolation --no-deps .

Optional VS Code configuration

Add the repository path to .vscode/settings.json if Pylance cannot resolve the local package:

{
  "python.analysis.extraPaths": [
    "/path/to/AI2Pot"
  ]
}

LAMMPS Interface

Install AI2Pot-lammps

1. Copy the AI2Pot interface files into the LAMMPS source directory:

$ cp AI2Pot/interface/lammps/AI2POT lammps/src
$ cp AI2Pot/interface/lammps/Makefile.mpi lammps/src/MAKE/

2. Build LAMMPS with the AI2POT package:

$ cd lammps/src
$ make yes-AI2POT
$ make -j 16 mpi \
    TORCH_ROOT=$(python -c "import torch, os; print(os.path.dirname(torch.__file__))") \
    AI2POT_ROOT=$(python -c "import ai2pot, os; print(os.path.dirname(ai2pot.__file__))")

AI2Pot pair style usage

pair_style ai2pot ./ai2pot_libtorch.pt
pair_coeff * * Ge Sb Te

Citation

[Liu2026] Hanyu Liu, Linggang Zhu, Xuanguang Zhang, Ning Yang, Jian Zhou, Zhimei Sun. AI2Pot: A scalable and unified framework for machine-learning interatomic potential development and large-scale molecular dynamics simulations, arXiv preprint arXiv:2607.06969 (2026).