Adding api-inference-community to huggingface_hub.

[Narsil]

• The idea is to make all external contributions centralized.
• Moved the quality checks (could make them uniform later)
• api-inference-community should be relatively independant from huggingface_hub still.

[julien-c]

Thanks! Looks good to me but adding @LysandreJik and @n1t0 to the party

[julien-c]

Does the CI take a long time? If it does, maybe we'll want to restrict the Inference-API's CI to changes to that folder.

[LysandreJik]

Cool, looking forward to having everything centralized.

[Narsil]

Not sure how easy to run conditional actions but if it's easy then yes let's do it !

Otherwise the CI is super fast as we only run the package CI, everything else is considered "slow" as it would require model downloads (and I don't want to force users to test against dummy models as it is a bit more complex to do).

Before deployment, we run the test_dockers slow CI which actually build the container + run the test suite on those with a live model. We can do that framework per framework making it livable (running tests against all frameworks all the time seems a bit like a waste since we're relying on docker to make sure no unforeseen change happens within the API)

Conditional full docker tests would be the best so that PR creators can be alerted early if something doesn't fit.

[Narsil]

Hi @julien-c do you think the changes made are OK?

I added a simple docker test only for allennlp (afaik, I would need one for each framework, because we can't have multiple on clauses in a single yaml file).

Also I am not 100% sure about this docker because it does not seem to run on this PR, is that normal ? I though GH actions should run on current PR, but maybe I'm wrong.

[julien-c]

not sure why the CI suite doesn't run, maybe it will after we merge this first PR? or maybe add a pull_request: event clause with the same paths?

[LysandreJik]

LGTM, thanks @Narsil!

I'm pretty sure it's normal that the CI doesn't run as long as this isn't on master, we had the same issue on transformers a while back.