VISSL supports (10.2, 11.1) CUDA version for these docker containers for both the pip and conda environments. We provide a parameterized Dockerfile to build the image based on the build environment. The different configurations are identified by a freeform string that we call a build environment. We support latest pytorch version (1.9.1) in our docker container. If you wish to change the pytorch version, please modify the Dockerfile pytorch installation commands.

You can specify build environment with string:

• cu$CU_VERSION # pip
• cu$CU_VERSION-conda # conda

Examples:

• Pip environment: cu102
• Conda environment: c102-conda

See build_docker.sh for a full list of terms that are extracted from the build environment into parameters for the image build.

NOTE: You need to have docker installed on your system. Follow the instructions on docker website to install it, if you don't have docker already.

You can verify your docker installation is fine by running a docker test:

docker run hello-world

1. Clone VISSL repo

cd $HOME && git clone --recursive git@github.com:facebookresearch/vissl.git && cd $HOME/vissl/

2. Build the docker image

• For pip environment

image=cu101 ./build_docker.sh

• For conda environment

image=cu101-conda ./build_docker.sh

This will build the image for the above permutation and then we can test this image

6. Test the image

run docker images and get the IMAGE_ID for the image we just built.

docker run --gpus all -it \
    --shm-size=8gb --env="DISPLAY" --volume="/tmp/.X11-unix:/tmp/.X11-unix:rw" ${IMAGE_ID}