Adding push_to_hub to CrossEncoder

[imvladikon]

Hi,
I've noticed that many new cool features have recently been added, so first of all, thank you for your work!

Description

The great CrossEncoder class currently lacks support for auto-uploading to huggingface through the push_to_hub function(although it's achievable using the transformers models API directly)
This PR adds inheritance from PushToHubMixin, which is the standard mechanism for performing push_to_hub in the Transformers library.

As a result, it's possible to do it:

from sentence_transformers import CrossEncoder

model = CrossEncoder("./mpnet-base-nli-matryoshka")
repo_id = "imvladikon/mpnet-base-nli-matryoshka"
model.push_to_hub(repo_id)

The resulting repo is here: https://huggingface.co/imvladikon/mpnet-base-nli-matryoshka where the README.md is generated automatically (which is the behavior that we have in the transformers library)

[tomaarsen]

Thanks a bunch for setting this up! I expanded it a bit, e.g. making some of the kwargs explicit to help with IDE auto-suggestions and by always setting the cross-encoder tag on the generated README (although the README is still quite empty).

• Tom Aarsen