Replace accelerator.use_fp16 in examples

[hlky]

What does this PR do?

examples_torch tests are currently failing due to usage of accelerator.use_fp16, use_fp16 was removed in accelerate#3098 and transformers CI uses accelerate@main so the property is no longer available.

This PR replaces accelerator.use_fp16 with accelerator.mixed_precision != "no" which is equivalent to the removed use_fp16 property.

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?

cc @muellerzr

[SunMarc]

Thx for fixing !

[SunMarc]

If we are using fp8, we need to pad to 16. Since these are simple examples + did not take into account fp8 from the start, I guess we can leave it like that. Otherwise, we would have to do the following

  if accelerator.mixed_precision == "fp8":
            pad_to_multiple_of = 16
        elif accelerator.mixed_precision != "no":
            pad_to_multiple_of = 8
        else:
            pad_to_multiple_of = None

[hlky]

Thanks for the review, I've gone ahead and added pad_to_multiple_of = 16 for fp8.

[ArthurZucker]

You know a lot more than me about this one @SunMarc ! Juste wondering when was mixed_precision introduced? (before 1.0?)

[hlky]

Looks like mixed_precision was introduced a while ago in accelerate#247, the now removed use_fp16 property was changed to use mixed_precision at the same time.

[SunMarc]

Looks like mixed_precision was introduced a while ago in huggingface/accelerate#247, the now removed use_fp16 property was changed to use mixed_precision at the same time.

Thanks for looking into ! Looks like this was done in 2022, so with <= accelerate v.0.20. The min version of accelerate in transformers is v0.26.0.

Merging this since this should be safe to do

[ArthurZucker]

Thanks both for checking!