Installation

Installation

This page covers the v2.0 and v2.1 installation workflow.

Dependencies

PAIPAI requires:

• CMake
• a C++17 compiler
• Python 3
• MaterialsFramework and its MLIP dependencies
• optional CUDA/cuDNN support for GPU runs

Before running PAIPAI workers, verify that the Python environment can import MaterialsFramework:

python -c "import materialsframework"

Clone

git clone https://github.com/siyazhu/PAIPAI.git
cd PAIPAI

Use the prefast development branch:

git checkout v2.1-dev

Use the v2.0 development branch:

git checkout v2.0-dev

Build and Install

mkdir build
cd build
cmake .. -DCMAKE_INSTALL_PREFIX=$HOME
cmake --build .
cmake --install .

The install step provides:

$HOME/bin/paipai
$HOME/bin/mc_paipai
$HOME/bin/findinter
$HOME/bin/fast_worker.py
$HOME/bin/slow_worker.py
$HOME/bin/warrencowley
$HOME/bin/packmc

Add the install directory to your PATH if needed:

export PATH="$HOME/bin:$PATH"

Python Used by Workers

The launcher uses the Python executable detected by CMake. You can override it at runtime:

export PAIPAI_PYTHON="$CONDA_PREFIX/bin/python"

This is useful on clusters where the installed launcher should use a specific conda environment.

GPU Environment

For GPU runs:

paipai --device cuda --ngpu 4 ...

The launcher distributes workers across GPUs using CUDA_VISIBLE_DEVICES.