1. 使用yml文件打包
conda activate your_env
conda env export > environment.yml
编写cond.def文件
Bootstrap: dockerFrom: continuumio/miniconda3%filesenvironment.yml%post/opt/conda/bin/conda env create -f environment.yml%runscriptexec /opt/conda/envs/$(head -n 1 environment.yml | cut -f 2 -d ' ')/bin/"$@"
生成镜像:
singularity build conda.sif conda.def
2. 利用tar包
2.1 安装conda-pack
pip install conda-pack
版本需要0.7以上。
2.2 导出tar包
conda-pack -n <MY_ENV> -o packed_environment.tar.gz
编写conda.def文件:
Bootstrap: dockerFrom: continuumio/miniconda3%filespacked_environment.tar.gz /packed_environment.tar.gz%posttar xvzf /packed_environment.tar.gz -C /opt/condaconda-unpackrm /packed_environment.tar.gz
生成镜像:
singularity build --fakeroot <OUTPUT_CONTAINER.sif> conda.def
3. 在已有基础上构建
def:
Bootstrap: localimage
From: local_image.sif%environment# set up environment for when using the container. /opt/conda/etc/profile.d/conda.shconda activate%postapt-get update -yapt-get install -y \build-essential \wget \cmake \g++ \r-base-dev \makeR -e "install.packages('cowsay', dependencies=TRUE, repos='http://cran.rstudio.com/')"# install minicondawget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.shbash Miniconda3-latest-Linux-x86_64.sh -b -f -p /opt/condarm Miniconda3-latest-Linux-x86_64.sh# install conda components - add the packages you need here. /opt/conda/etc/profile.d/conda.shconda activateconda install -y -c conda-forge numpy cowpyconda update --all4. 沙盒模式
4.1 构建沙河目录
singularity build --sandbox lolcow/ library://sylabs-jms/testing/lolcow
4.2 进入沙盒
singularity shell --writable lolcow/
4.3 将沙盒打包成sif
singularity build lolcow.sif lolcow/
5. 设置环境变量
pytorchcmake未设置cuda环境变量
SET(CMAKE_INCLUDE_PATH ${CMAKE_INCLUDE_PATH} "path\\boost_1_80_0")
SET(CMAKE_LIBRARY_PATH ${CMAKE_LIBRARY_PATH} "path\\boost_1_80_0\\libs")
可以通过如下设置:
%environmentexport CUDA_INCLUDE_DIRS=/opt/conda/cuda/includeexport CUDA_CUDART_LIBRARY=/opt/conda/cuda/libexport LIBRARY_PATH=/opt/conda/cuda/lib:$LIBRARY_PATHexport CPATH=/opt/conda/cuda/include:$CPATHexport PATH=/opt/conda/cuda:$PATH
%postmkdir -p /opt/conda/cudaconda install cuda -c nvidia -p /opt/conda/cudamkdir -p /opt/conda/cudnnconda install -c anaconda cudnn -p /opt/conda/cudnnexport PATH=/opt/conda/cuda:$PATH
)
