The default docker container assumes you have a CUDA enabled device - this affects both the base docker image I start from (pytorch:1.11.0-cuda11.3-cudnn8-devel) and that I pass all GPUs in the launch scripts via --gpus all.
make run launches imagename
make build builds Dockerfile as an image called "imagename"
make jupyter launches jupyter-lab
-
Use this repo as a template
-
git grep imagename, change imagename to a new name, or ideally, your image name hosted on dockerhub -
change the setup.py file project name so something of your own
Go here.
Go to dockerhub. This is useful so that others do not need to rebuild the container again with different versions.