Distrubuted DNN Training on Heterogeneous GPUs
创建编译目录
mkdir build && cd build
生成makefile
cmake ..
支持的参数
-DBUILD_STATIC_LIB=ON # 开启静态库编译
编译
make
python包(可选)
pip install build
python -m build
pip install dist/xxx.whl
sudo apt install openmpi-bin openmpi-common libopenmpi-dev
https://docs.anaconda.com/miniconda/
conda create -n py310 python=3.10
pip3 install torch torchvision torchaudio
python -c "import torch; print(torch.cuda.is_available())"
sudo apt-get install libgoogle-glog-dev
git submodule update --init --recursive