Enabling Keras download stats

[merveenoyan]

This PR adds library_name and library_version to snapshot_download in Keras mixin.

I think existing from_pretrained_keras tests should cover this so I didn't write an additional test.

[HuggingFaceDocBuilderDev]

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

[merveenoyan]

I'm checking out why test_filter_models_with_complex_query and test_tags are failing.

[adrinjalali]

What is the consequence of this change?

[merveenoyan]

(Leaving notes to myself)
I don't think it's related to my PR honestly but:

(Pdb) api.get_model_tags()
*** KeyError: 'language'

(called from __init__ of ModelSearchArguments() that is called inside the test)
but I do have "language" in the response returned from get_model_tags() 🧐

[merveenoyan]

@adrinjalali when we add library name and version to snapshot download it sends a request with the info inside the header that later gives us statistics about Keras downloads.

[adrinjalali]

Ah I see, so it doesn't really change anything on the client side. Got it.

[merveenoyan]

There's a problem with dataset tags. I'm fixing it.

[merveenoyan]

There's a dataset that is empty after the dataset batterypapers, (this is the dict of that dataset
{'id': 'dataset:', 'label': '', 'type': 'dataset'},).
It breaks _unpack_and_assign_dictionary() which causes a bug in the tests (and initialization of DatasetTags class in general).
To whom can I talk to? Maybe @osanseviero and @lhoestq.

[adrinjalali]

Should we be ignoring those datasets then? As in, if there's an issue creating the class, ignore the entry. In general, we should be safe against malformed data that people put on the hub.

[merveenoyan]

@adrinjalali if @osanseviero approves. Zach said he reached out to owners of dataset to remove them.

[osanseviero]

This internal PR from @julien-c will prevent this situation in the future https://github.com/huggingface/moon-landing/pull/2822

[merveenoyan]

@osanseviero I re-ran and problems seem to be gone.

[osanseviero]

Yes, we chatted with the model authors and fixed in the backend. Thanks!

[osanseviero]

Looks good! Thanks 🔥