Step by step guide on adding Model Hub support to libraries

[LysandreJik]

This PR adds a detailed guide on adding Model Hub support to downstream libraries.

It covers:

• Downstream: Usage from the hub
• Upstream: Repository creation/uploads to the hub
• Setting up the tool stack: inference API, widgets

Pinging current library owners that have integrated with the model hub for potential feedback 🤗:

• Adapter-transformers: @calpt
• AllenNLP: @epwalsh @dirkgr
• Asteroid: @mpariente @jonashaag
• ESPNet: @sw005320 @kan-bayashi
• Flair: @alanakbik
• Pyannote: @hbredin
• Sentence-Transformers: @nreimers
• SpaCy: @ines
• TensorFlowTTS: @dathudeptrai
• timm: @rwightman

Looking forward to your feedback! The file is visible here, and it will end up on the huggingface.co website when merged.

[Narsil]

Very nice ! A thorough and easy to read doc item IMO

[osanseviero]

Looks great, neat and easy to read!

I left minor nits.

Just three suggestions

• Use Hugging Face consistently (sometimes you use HuggingFace and sometimes HuggingFace)
• Similarly, use either the Hub or Hugging Face Hub consistently. turbo nit: please capitalize it :)
• If possible, add a link to a PR that shows how to do each of these three parts (download example, etc)

[nreimers]

Nice documentation :)

In the future, the documentation on adding the docker image to the inference API could be a bit more extensive, e.g. what is the principle architecture, what is the purpose of the pipelines, how to add different tasks that a library is supporting etc.

So far for me it would be mainly checking the other examples and copy&pasting their code and hoping that I don't break anything.

Also how to test your docker container locally if everything is correct (i.e. how does the input request has to look like, and are there requirements for the output?)

Not super important for now, but might be interesting in the long term if we want to include more libs in the inference API

[mpariente]

Thanks for pinging, this looks great.

[sw005320]

Year, that looks great!

[epwalsh]

This is great @LysandreJik, super helpful!

[osanseviero]

I did a second pass, updated some links and added some new images. Feel free to take a look :)

[julien-c]

lgtm