DeePMD-kit安装实战：嘉庚超算

嘉庚超算中心没有统一安装DeepMD-kit软件，用户使用前需要自行编译。本文参考最佳实践，基于嘉庚超算预装的模块进行。此处以DeepMD-kit v2.2.0版本为例。

初次安装

1. 创建虚拟环境（此处以deepmd为例）

   module load anaconda/2020.3
   conda create -n deepmd python=3.9
   

2. （可选）虚拟环境激活/退出的配置，也可将activate.sh中代码每次手动设置

   # replace your own username here!
   mkdir -p $CONDA_PREFIX/etc/conda/activate.d
   touch $CONDA_PREFIX/etc/conda/activate.d/activate.sh
   mkdir -p $CONDA_PREFIX/etc/conda/deactivate.d
   touch $CONDA_PREFIX/etc/conda/deactivate.d/deactivate.sh
   conda env config vars set LD_LIBRARY_PATH=$tensorflow_root/lib:$deepmd_root/lib:$CONDA_PREFIX/lib:$LD_LIBRARY_PATH
   

   • $CONDA_PREFIX/etc/conda/activate.d/activate.sh

   module load intel/2018.3
   module load gcc/9.2
   module load cmake/3.21
   module load cuda/11.3
   module load lammps/2022.6.23
   
   export CC=`which gcc`
   export CXX=`which g++`
   export FC=`which gfortran`
   
   # replace CONDA_PREFIX and deepmd_source_dir!!!
   export deepmd_source_dir=/public/home/username/apps/deepmd-2.2.0
   export tensorflow_root=$deepmd_source_dir/_skbuild/tensorflow_root
   export deepmd_root=$deepmd_source_dir/_skbuild/deepmd_root
   export LAMMPS_PLUGIN_PATH=$deepmd_root/lib/deepmd_lmp
   

   • $CONDA_PREFIX/etc/conda/deactivate.d/deactivate.sh

   module unload intel/2018.3
   module unload gcc/9.2
   module unload cmake/3.21
   module unload cuda/11.3
   module unload lammps/2022.6.23
   
   unset deepmd_source_dir
   unset tensorflow_root
   unset deepmd_root
   unset LAMMPS_PLUGIN_PATH
   

   设置好后，重启虚拟环境。此后每次激活虚拟环境时，会自动加载相应的模块。 3. 训练代码安装

   pip install tensorflow==2.7 --upgrade
   pip install scikit-build ninja
   pip install protobuf==3.20
   cd $deepmd_source_dir
   export DP_VARIANT=cuda
   pip install .
   

   4. (可选)第三方接口安装

   mkdir -p $tensorflow_root/lib 
   cd $tensorflow_root
   ln -s $CONDA_PREFIX/lib/python3.9/site-packages/tensorflow/include .
   cd lib
   ln -s $CONDA_PREFIX/lib/python3.9/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so libtensorflow_cc.so
   ln -s $CONDA_PREFIX/lib/python3.9/site-packages/tensorflow/libtensorflow_framework.so.2 .
   ln -s libtensorflow_framework.so.2 libtensorflow_framework.so
   
   mkdir -p $deepmd_source_dir/source/build
   mkdir -p $deepmd_root
   cd $deepmd_source_dir/source/build
   cmake -DLAMMPS_SOURCE_ROOT=/public/software/lammps/lammps-2022.6.23-intel -DUSE_TF_PYTHON_LIBS=TRUE -DUSE_CUDA_TOOLKIT=TRUE -DTENSORFLOW_ROOT=$tensorflow_root -DCMAKE_INSTALL_PREFIX=$deepmd_root ..
   make -j20
   make install
   

代码更新

1. Python代码

   cd $deepmd_source_dir
   export DP_VARIANT=cuda
   pip install .
   

2. C++代码

   cd $deepmd_source_dir/source/build
   make -j20
   make install
   

评论