Fix TensorFlow pakage check

[jayfurmanek]

What does this PR do?

This fixes the _is_package_available check for tensorflow variants (like tensorflow-rocm) where their meta-data differs from the package name. There is an list of available candidate names there, but they are never used since _is_package_available will return false preventing the candidate list from being checked.
This also adds tf-nightly-rocm to the list of candidates.

Fixes # (issue)
N/A

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

This fixes a small functional error, no documentation updates are required.

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

• tensorflow review: @gante and @Rocketknight1

[Rocketknight1]

Taking this one!

[Rocketknight1]

Hey @jayfurmanek, I checked this out! Adding tf-nightly-rocm to the candidate list is appreciated, but I don't think the other changes are required, and they seem to break a lot of the codebase! Can we revert that part of the PR and just keep the package name addition, or did that not work for you?

Update: Part of the issue might be that importlib.metadata is typoed as importlib_metadata - but I think we still should be okay if we just revert that change!

[jayfurmanek]

The problem with the check as-is is that it does _is_package_available() with "tensorflow" and then does the meta data check there as well. For tensorflow-rocm that part will fail and _is_available will be marked as False, so the candidates list below is skipped altogether.

[Rocketknight1]

Hi @jayfurmanek I just tested locally and installing with pip install tensorflow-rocm works fine, and causes transformers.is_tf_available() to correctly return True. Can you tell me where I can get a wheel for tf-nightly-rocm so I can test that?

[jayfurmanek]

Try: http://ml-ci.amd.com:21096/job/tensorflow/job/nightly-rocmfork-develop-upstream/job/nightly-build-whl/

[Rocketknight1]

@jayfurmanek I confirmed after installing one of those wheels that just adding tf-nightly-rocm to the list is sufficient! Can we revert the rest of the PR and just keep that bit?

[Rocketknight1]

If you want to test it yourself, you can run transformers.is_tf_available() to see if transformers can detect a TF installation. It works fine for me once the list is updated, with no other changes.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[jayfurmanek]

I was using is_package_available() for tf not is_tf_available. With is_tf_available() it works.
I still think is_package_available() is a bit wrong though.
PR updated.

thx

[Rocketknight1]

Yeah, that code is a little wonky, and we'll have to do a proper refactor eventually. I think this resolves the issue here for now, though, so thanks for the PR!