Avoid using torch's Tensor or PIL's Image in chat template utils if not available

[RezaRahemtola]

What does this PR do?

This PR adds conditional checks to the types mapping definitions to avoid defining types that are conditionally imported previously, if the same check doesn't succeed.

This allows a user (like me) to use transformers for simple stuff like tokenizers without needed to install PIL and/or torch if not needed (I'm not using the tokenizers library directly as it doesn't include chat templating functions, and I'm not using torch/tensorflow/JAX as I'm just calling models run locally with llamacpp, installing one of those just for chat templating seems overkill).

Before, a simple call to tokenizer.apply_chat_template would result in this error:

  File "...../venv/lib/python3.11/site-packages/transformers/utils/chat_template_utils.py", line 79, in _get_json_schema_type
    Image: {"type": "image"},
    ^^^^^
NameError: name 'Image' is not defined

With this fix, there's not error: if PIL (or torch) isn't installed, it's not imported (already the case) and not used (added here).

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?

Who can review?

• tokenizers: @ArthurZucker
• chat templates: @Rocketknight1

[LysandreJik]

Also cc @zucchini-nlp

[Rocketknight1]

Yes, this looks good to me! For clarity, this function is called when auto-generating JSON schema for functions passed as tools to the chat template interface.

The main thing to be wary of here is that this change might cause slightly different behaviour on systems where Image and Tensor types are unavailable. However, I think we can safely assume that when the user has defined a function with Image or Tensor type hints in the header, that means they have the relevant libraries.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[Rocketknight1]

@zucchini-nlp @LysandreJik this PR is quite straightforward, and I think it's okay to accept - are you okay with me merging it once tests pass?

[zucchini-nlp]

I am not qualified for review here, so whatever you find reasonable @Rocketknight1 :)

[Rocketknight1]

In that case, since this is a simple bugfix, I think it's okay to merge. One last step, though - @RezaRahemtola can you fix up the code quality tests? To do that:

1) pip install transformers[quality]
2) make style in the transformers directory
3) commit and push

Never mind - this is actually a CI issue. Rerunning it, you don't need to do anything @RezaRahemtola !

[LysandreJik]

Let's go!

[Rocketknight1]

Documentation build is failing but this PR doesn't touch the docs at all, so I'm assuming that's an issue with the doc builder. Merging, and thanks @RezaRahemtola for the PR!