Improve generate docstring (for TF and FLAX)

[JoaoLages]

What does this PR do?

Just a continuation PR of #18198 for TF and FLAX code

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).

Who can review?

@sgugger @gante

[sgugger]

Thanks for expanding to the other generate functions!

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[sgugger]

Looks like we just need a quick make style to be ready to merge :-)

[sgugger]

Mmm no the formatting has done something very wrong here, we can't have that. There is likely some syntax error in the docstring.
Problem looks to be in the input_ids argument at line 422 of the TF generation files, the type should all be on one line.

[JoaoLages]

Mmm no the formatting has done something very wrong here, we can't have that. There is likely some syntax error in the docstring. Problem looks to be in the input_ids argument at line 422 of the TF generation files, the type should all be on one line.

make extra_style_checks does that automatically. What do you propose?

[sgugger]

As I said, there is a syntax error in the docstring that makes the styling script behave erratically. The first step is to revert the changes, fix the syntax error then re-run it.

[JoaoLages]

I changed this line

 input_ids (`tf.Tensor` of shape `(batch_size, sequence_length)`, `(batch_size, sequence_length,
            feature_dim)` or `(batch_size, num_channels, height, width)`, *optional*):

to this

 input_ids (`tf.Tensor` of shape `(batch_size, sequence_length)`, `(batch_size, sequence_length, feature_dim)` or `(batch_size, num_channels, height, width)`, *optional*):

but then make extra_style_checks reverts the change

[sgugger]

Ah yes, just tried locally and it's due to the empty line between Parameters: and input_ids. If you remove it, then your changes should not be overwritten.

[JoaoLages]

Ah yes, just tried locally and it's due to the empty line between Parameters: and input_ids. If you remove it, then your changes should not be overwritten.

Nice catch, but it still does look strange in the docs 🤔

[sgugger]

You did not revert the bad changes made by the style doc string, that's why. I showed you what needs to happen on the first arguments (basically the type has to be on its own line ending with a colon) but the easiest is probably to start from the commit before the style changes.

[gante]

Uhmm, there is something wrong with the automatic styler -- e.g. the pytorch generate file should not be touched at all in this PR. As Sylvain wrote, the easiest solution is to start from a new branch 🤔

[JoaoLages]

Uhmm, there is something wrong with the automatic styler -- e.g. the pytorch generate file should not be touched at all in this PR. As Sylvain wrote, the easiest solution is to start from a new branch 🤔

The docs seem fine now, right? It's just that this test, doc-builder style src/transformers docs/source --max_len 119 --check_only --path_to_docs docs/source, is not allowing to have more than 119 characters per line, but we need it here.

[sgugger]

Nope, the docs are not fine for the PyTorch side with all the changes in this PR (and as @gante mentioned that file should not be touched at all). The doc-style is completely comfortable with lines that are more than 119 chars when it identifies they are parameter introduction lines, you just needed to remove the blank line between Parameters: and the first argument in generate.

[JoaoLages]

Ah right, it has these strange hyphens...

I will close the PR then, let's disregard these changes.