Specify CUDA information within docker for GPU tests

[seanpmorgan]

Looks like the GPU tests were truly hanging:
https://source.cloud.google.com/results/invocations/67100f67-5a71-431b-952a-4efb61e8d1b8/targets/tensorflow_addons%2Fubuntu%2Fgpu%2Fpy3%2Fcontinuous/log

Interested to know why the release call to build_and_release somehow picked up default values:
https://github.com/tensorflow/addons/blob/master/tools/releases/release_linux.sh#L40
https://github.com/tensorflow/addons/runs/575305456?check_suite_focus=true#step:6:433

While the same call in the Dockerfile hangs waiting for a response:
https://github.com/tensorflow/addons/blob/master/tools/docker/gpu_tests.Dockerfile#L15

[gabrieldemarmiesse]

Interested to know why the release call to build_and_release somehow picked up default values

It seems that in github actions, (maybe because it's in a build step), user input is not available, as such, in this line: https://github.com/tensorflow/addons/blob/master/configure.py#L174

answer is "" (empty string).

In google cloud, for some reason, (maybe the -t argument in the docker run), python thinks that user input is available, as such it doesn't return en EOF in this line: https://github.com/tensorflow/addons/blob/master/configure.py#L51 .

I believe that we shouldn't ask for user input and always use the environement variables or default values. I'll make a PR for that.

[gabrieldemarmiesse]

I think this can be closed because it's superseded by #1634