Docker image fails to build due to bad cuda repo pki

[qhaas]

The base image of pytorch/pytorch:1.4-cuda10.1-cudnn7-devel is ~2 years old and it appears that the nvidia apt repository leveraged by it has had its pkis revoked or they have expired, resulting in the inability to build this image (see attached docker build log) .

This can be fixed in a number of ways

• Update to a newer versions of a GPU enabled variant of pytorch/pytorch for the base image (big change)
• Disable the cuda repo by adding a RUN sed -i 's|^deb|#deb|' /etc/apt/sources.list.d/cuda.list before the first RUN (least invasive since packages from this repo do not seem to be deployed in the Dockerfile)
• Update the PKIs by adding a RUN apt-get clean && apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/3bf863cc.pub before the first RUN (this will break again once the PKIs loaded by it expire)