Uniformize hf_api a bit and add support for Spaces

[julien-c]

No description provided.

[adrinjalali]

Some of the changes here are breaking changes. I think we should introduce a deprecation cycle like the other PRs if we're breaking API.

[julien-c]

@adrinjalali I think only two lines in this diff are potential breaking changes, so I'll revert them. Other than that i don't think there any BC, but I'll check

[LysandreJik]

Weird that CI didn't trigger on your PR

[julien-c]

@LysandreJik probably GitHub being a bit down

[LysandreJik]

Other than the sha addition as a kwarg to DatasetInfo to correspond better to the ModelInfo, I don't see a breaking change; did I miss anything @adrinjalali?

[adrinjalali]

We also need to document the new methods in the README here.

[LysandreJik]

Added documentation for the methods that weren't touched by #782 so that everything is in the same format.

[julien-c]

BTW feel free to rebase this on main if you'd like!

[LysandreJik]

Yes, will now rebase on main as the big doc PR was just merged.

[HuggingFaceDocBuilderDev]

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

[adrinjalali]

Should the new methods also be included in a user guide?

[LysandreJik]

I should have addressed all comments.

Should the new methods also be included in a user guide?

They'll be in the API reference on merge, which I think is a good first step.

[LysandreJik]

Need to patch the failure.

[LysandreJik]

The switch from HfFolder.get_token() to _validate_or_retrieve_token in the model_info results in a bunch of failing tests. Taking a step back, I think it would be better to revert it as out of scope for this PR (it currently is like that on main), and address in a specifc PR.

@adrinjalali, please let me know if that works for you, and if so, if we can merge this PR.

[adrinjalali]

mostly cosmetic comments, except HfFolder.get_token(). I think if a new method is failing on using validate_or_retrieve_token, it's in the scope of the same PR to see why it'd fail. We have 9 instances of using _validate_or_retrieve_token and only two HfFolder.get_token(). We don't have to remove the existing ones here, but new code shouldn't be use it.

[LysandreJik]

Thanks for your feedback. I've taken care of all your comments, except regarding the _validate_or_retrieve_token, as I stand by what I said in the previous comment. I understand how

I think if a new method is failing on using _validate_or_retrieve_token, it's in the scope of the same PR to see why it'd fail.

that would be sensible, but the two methods are not designed to work together. The _validate_or_retrieve_token method exists to either validate the passed token, or retrieve the logged-in token, before validating it. It cannot be used as a no-op in case the passed token is None and the user isn't logged-in.

That's exactly what model_info (from which a direct copy is made for both dataset_info and space_info) is advertised to do, though:

Get info on one specific model on huggingface.co

Model can be private if you pass an acceptable token or are logged in.

In the case where no acceptable token is passed and the user isn't logged-in, then it should still work. Updating to use _validate_or_retrieve_token breaks that, as it raises an error.

Furthermore, I think that since model_info aims to be the exact same as dataset_info and space_info, but for different repository types, they should not use different approaches to retrieve the token, as situations like the one above can arise, which results in a desynchronisation of the three methods and an API behaving a bit differently acorss seemingly identical methods. I think in this case, if it is out of scope to refactor model_info, then dataset_info and space_info ought to be designed exactly like model_info is on main.

[LysandreJik]

Rebasing on main to fix the tests.

[adrinjalali]

Ok, opened #837 to discuss the token issue. Happy to merge this as is and fix that issue later.