Pin Docker image for testing on GPUs

[akihironitta]

What does this PR do?

Temporary fix for #12314. For unblocking other PRs, tries to pin the most recent docker image in which the tests were running successfully.

Specifically, the one used in https://dev.azure.com/PytorchLightning/pytorch-lightning/_build/results?buildId=61012&view=logs&j=3afc50db-e620-5b81-6016-870a6976ad29&t=bd0a18db-4e66-438b-93e9-00d86028355e.

Does your PR introduce any breaking changes? If yes, please list them.

None

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• [n/a] Did you make sure to update the documentation with your changes? (if necessary)
• [n/a] Did you write any new necessary tests? (not for typos and docs)
• [n/a] Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• [n/a] Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

cc @tchaton @rohitgr7 @akihironitta @carmocca @Borda

[akihironitta]

The CI job hasn't been finished yet, but it seems alright so far. https://dev.azure.com/PytorchLightning/pytorch-lightning/_build/results?buildId=61478&view=logs&j=3afc50db-e620-5b81-6016-870a6976ad29

[daniellepintz]

Thanks @akihironitta!

[Borda]

also curious how we uploaded corrupted image as each image at the end shall have its own testing...

[akihironitta]

@Borda The corrupted image doesn't raise any error as seen in this "successful run" build-CUDA (3.7, 1.8) for instance. Building itself was successful, but since there's no sanity check (import horovod.torch raises a warning but not an exception) at the moment in the Dockerfile, the built image got uploaded to the hub. As you've already done in your PR, something like from horovod.torch import nccl_built; nccl_built() should avoid this issue since it raises an exception if the dependency isn't installed in the image.

[awaelchli]

Thank you so much @akihironitta