Skip to content

PySAGES installation notes

Jump to bottom
Yiheng Wu edited this page May 7, 2024 · 3 revisions

Collections of recipes for installing PySAGES.

Installation of HOOMD and PySAGES using a conda environment in a PC

Requirements:

  • A git installation
  • A conda installation (consider using miniforge instead of miniconda)
  • CUDA (optional) if you have a Nvidia GPU

Steps:

  1. Create new conda environment 
    conda create -n pysages
  2. Activate your conda environment 
    conda activate pysages
  3. Install hoomd 
    conda install -c conda-forge hoomd
    or if you have a GPU 
    conda install -c conda-forge hoomd=*=*gpu*
  4. Get and build hoomd-dlext 
    git clone https://github.com/SSAGESLabs/hoomd-dlext
cd hoomd-dlext
./build.sh

Installation of HOOMD and PySAGES on midway2

Load required modules:

module load python/anaconda-2021.05 
module load gcc/10.2.0 cmake/3.15 cuda/11.2 eigen/3.3

Steps:

  1. Create a conda environment

    conda create -n pysages
    source activate pysages

  2. Install HOOMD required libraries

    conda install -c conda-forge pybind11
    conda install python=3.8 
#[Downgrade for numba. If pybind11 installs python=3.10; numba installation may fail.]

  3. Download HOOMD source

    curl -O https://glotzerlab.engin.umich.edu/Downloads/hoomd/hoomd-v2.9.7.tar.gz
tar -zxvf hoomd-v2.9.7.tar.gz
cd hoomd-v2.9.7.tar.gz
mkdir build
cd build

  4. Install HOOMD

    PYTHON_EXECUTABLE=$(which python3)
HOOMD_ROOT=$(${PYTHON_EXECUTABLE} -c 'import site; print(site.getsitepackages()[0])')`
cmake ../ \
    -DCMAKE_INSTALL_PREFIX=$HOOMD_ROOT \
    -DCOPY_HEADERS=ON \
    -DENABLE_CUDA=ON  \
    -DCMAKE_C_COMPILER=/software/gcc-10.2.0-el7-x86_64/bin/gcc\
    -DCMAKE_CXX_COMPILER=/software/gcc-10.2.0-el7-x86_64/bin/g++
make -j 10 # On an interactive GPU node.
make install

  5. Install HOOMD-dlext [currently minimum version is 3.16 (can be installed using source code or conda environment easily); midway2 max version is 3.15]

    git clone https://github.com/SSAGESLabs/hoomd-dlext
source build.sh # Edit cuda path to cuda/11.2 on midway2

  6. Install other required libraries and JAX

    conda install numpy numba
    conda install -c conda-forge cupy
    conda install gsd matplotlib
    pip install --upgrade "jax[cuda11_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

  7. Install PySAGES

    git clone pysages
cd to pysages
pip install .

Installation of OpenMM and PySAGES using a conda environment on a Mac

Requirements:

  • A git installation
  • A conda installation

Steps:

  1. Create new conda environment 
    conda create -n pysages python=3
  2. Activate your conda environment 
    conda activate pysages
  3. Install OpenMM 
    conda install -c conda-forge openmm
  4. Install other required libraries and JAX 
    conda install -c conda-forge clang_osx-64
conda install -c conda-forge clangxx_osx-64
conda install -c conda-forge swig
conda install -c conda-forge plum-dispatch
conda install -c fastai fastcore
conda install -c conda-forge numba
pip install jax  
pip install jaxlib
pip install --upgrade jaxlib
  5. Install openmm-dlext 
    git clone https://github.com/pabloferz/openmm-dlext
cd openmm-dlext
cmake -S . -B build
cmake --build build --target install
  6. Install PySAGES 
    git clone https://github.com/SSAGESLabs/PySAGES.git
cd  PySAGES
pip install .

Installation of OpenMM and PySAGES on midway2

Load required modules:

module load python/intel-2020.up1
module load gcc/10.2.0 cuda/11.2

Steps:

  1. Create a conda environment

    conda create -n pysages-openmm
    source activate pysages-openmm

  2. Install OpenMM using conda

    conda install -c omnia openmm

  3. Install jaxlib and jax cuda with pip

    pip install jaxlib
    pip install --upgrade "jax[cuda11_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

  4. Download and install cmake/3.16 (may take few min)

    wget https://cmake.org/files/v3.16/cmake-3.16.0.tar.gz
tar -zxvf cmake-3.16.0.tar.gz
mkdir installed-files
cmakeLocation=$(pwd)
./configure --prefix=$cmakeLocation/installed-files
make install

  5. Install OpenMM-dlext [currently minimum version is 3.16]

    git clone https://github.com/SSAGESLabs/openmm-dlext.git
$cmakeLocation/installed-files/bin/cmake -S . -B build
$cmakeLocation/installed-files/bin/cmake --build build --target install

  6. Install PySAGES

    git clone pysages
cd to pysages
pip install .

  7. If it doesn't work, upgrade python (if prompted, upgrade openmm), then rerun jaxlib

    conda install python=3.9.10
    pip install --upgrade "jax[cuda11_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

  8. Re-install openmm-dlext.

    cd openmm-dlext
rm -f -r build
$cmakeLocation/installed-files/bin/cmake -S . -B build
$cmakeLocation/installed-files/bin/cmake --build build --target install

  9. Re-install PySAGES

    cd pysages
pip install .

  10. install cupy

    conda install cupy

Installation of PySAGES on midway3

This is a summary of this discussion and this discussion.

Steps:

  1. Load the modules:

    module load python/anaconda-2021.05
module load pysages

  2. Then, if you want to use your own version of PySAGES, that environment can be cloned an modified with the following:

    conda create --clone pysages-0.4.0 --name mypysages   # clone the env
conda activate mypysages
cd PySAGES        # move to your local version of github.com/SSAGESLabs/PySAGES
pip install -e .  # run inside pysages directory to get latest dev in editable mode
pip install --upgrade jaxlib

  3. Then you are ready to submit PySAGES jobs.
    NOTE. In case you encounter a No GPU/TPU found, falling back to CPU error, upgrading both jax and jaxlib by using the following command (see here for details) should solve that:

    pip install --upgrade pip
### OLDER (seems not to work anymore)
### # Installs the wheel compatible with Cuda 11.4+ and cudnn 8.2+ (deprecated).
### pip install "jax[cuda11_cudnn82]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
# CUDA 11 installation (this works as of 2023-08-01)
# Note: wheels only available on linux.
pip install --upgrade "jax[cuda11_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
Clone this wiki locally