Provided a Dockerfile and usage instructions.

[brassel]

As discussed in issue #1 here is a Dockerfile with a much smaller dependency footprint. I have also simplified the interface for a quick trial, reducing the parameters to device only. Everyone wishing to tinker more deeply with your code will be able to adapt the Dockerfile, I am sure. I have included a Readme describing how to use the container for a quick test.

It might be worth noting that I set the base image to

nvcr.io/nvidia/nvhpc:23.5-devel-cuda_multi-ubuntu22.04

I think your original image was also from nvcr or was it not?

[jspenmar]

Hi,

Thanks for providing this PR, it looks good! Since the new dockerfile is only for the quickstart (e.g. it doesn't contain pytorch-lightning, wandb and such), maybe it's worth moving it to a new directory (e.g. docker_quickstart) rather than completely removing the existing one.

Kind regards, Jaime