feat(lang): Support default CUDA about envd HOT 6 CLOSED

We should choose the corresponding CUDNN and CUDA version according to PyTorch/TF version, I think.

Thus we may need to keep a compatibility table in the MIDI, then try to get the version of TF/Pytorch, then choose the corresponding version.

The syntax maybe

base(gpu = True)

WDYT @VoVAllen

from envd.

In the base case, we make CUDA a separate layer. When we need it we merge it into the base image.

But not sure if it is possible.

At least we keep a base image list in MIDI, and use different layer when choosing different CUDA version

from envd.

https://gitlab.com/nvidia/container-images/cuda/blob/master/dist/11.6.0/ubuntu2004/devel/Dockerfile

It's nvidia/cuda devel dockerfile.

https://github.com/tensorflow/tensorflow/blob/master/tensorflow/tools/dockerfiles/dockerfiles/devel-gpu.Dockerfile

It's tensorflow gpu devel dockerfile

from envd.

Currently I'm not sure how user would like to control over the cuda version. I think we can postpone the decision here to the config override part.
Also I don't think it's a good idea to determine cuda version based on PyTorch/TF version. We can specify it explicitly for now.

from envd.

SGTM

from envd.

Fixed in #42

from envd.