Add a condition for nested_detach

[haikuoxin]

What does this PR do?

def nested_detach(tensors):
    "Detach `tensors` (even if it's a nested list/tuple/dict of tensors)."
    if isinstance(tensors, (list, tuple)):
        return type(tensors)(nested_detach(t) for t in tensors)
    elif isinstance(tensors, Mapping):
        return type(tensors)({k: nested_detach(t) for k, t in tensors.items()})
    return tensors.detach() if isinstance(tensors, torch.Tensor) else tensors

The reason is that Trainer predict_step executes the nested_detach function to ensure that every value in inputs has been detached. One assumption here is that all model inputs have been converted to tensors.

However, the output of the dataset of some model projects is not a tensor, but a string. If the input strings is converted into a tensors through tokenizer in the forward function, it will be problematic to perform detach on the string. so adding a condition to fixing it.

Fixes #31852

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?

@muellerzr

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.

[muellerzr]

Thanks, makes sense to do this to me!

cc @SunMarc @amyeroberts

[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.

[amyeroberts]

Thanks for adding this!