❗ Need to build torch using GPU ❗
❗ Update: requires NVIDIA GPU + changing the Docker runtime. See https://stackoverflow.com/questions/59691207/docker-build-with-nvidia-runtime ❗
❗ The build requires quite a lot of space. It is recommended to build on a NVIDIA capable computer with more than 60GB of space. If working on a VM, please see https://evodify.com/change-docker-storage-location/ so that you can build on a larger volume than the base ❗
📝 Note: currently, the GitHub Action will not generate the container -- it is too big (~10.6GB) for the Action VMs. To build GPU support, requires Docker to be built with GPU
To build locally:
git clone https://github.com/cyverse-vice/jupyterlab-pytorch
cd latest
docker build -t jupyterlab-pytorch:22.01 .
test locally with GPU:
docker run -it --rm --gpus=all -p 8888:8888 -e REDIRECT_URL=http://localhost:8888 harbor.cyverse.org/vice/jupyter/pytorch:latest
test with CPU only:
docker run -it --rm -p 8888:8888 -e REDIRECT_URL=http://localhost:8888 harbor.cyverse.org/vice/jupyter/pytorch:latest
Open your browser and go to http://localhost:8888