Update metadata loading for oneformer

[amyeroberts]

What does this PR do?

Previously, the loading of the metadata file for oneformer was effectively hardcoded to download a file from the hub. This PR updates the prepare_metadata method to allow for loading of local files as well as model repos.

from transformers import OneformerImageProcessor

image_processor = OneformerImageProcessor(
    checkpoint, 
    repo_path="path/to/file", 
    class_info_file="local_metadata.json"
)

Fixes #23116 and partially addresses #27572

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?

[ydshieh]

2 nits but LGTM, thanks!

[ydshieh]

any reason to try without repo_type=dataset?

[amyeroberts]

Just that someone might store it under the model repo rather than the dataset. As the model is trained on a dataset - there's already a coupling and we already store data specific mappings e.g. id2label on the model side.

I can remove if you think it's better?

[ydshieh]

It's ok :-)

[ydshieh]

Could be more precise: if unset or the corresponding file could not be found locally ...

[amyeroberts]

If unset, then repo_path is set to "" - effectively making the filename class_info_file rather than repo_path/class_info, but it's independent of whether it can be found locally.

Would this be clearer:

            Path to hub repo or local directory containing the JSON file with class information for the dataset.
            If unset, will look for `class_info_file` in the current working directory.

And then adding an additional check in load_metadata:

def load_metadata(repo_id, class_info_file):
    fname = os.path.join("" if repo_id is None else repo_id, class_info_file)

    if not os.path.exists(fname) or not os.path.isfile(fname):
        if repo_id is None:
            raise ValueError(f"Could not file {fname} locally. repo_id must be defined if loading from the hub")

[ydshieh]

Either works for me, but we can go for the above one for simplicity.

[amyeroberts]

I did both in a777668 :)

[ArthurZucker]

Thanks fo the cleanup