♻️ Export individual functions, precise typing

[coyotte508]

cc @Grsmto @vvmnnnkv @TimMikeladze

Inference endpoints do not accept model anymore.

Also, individual functions for the various tasks are exported, taking an additional optional accessToken parameter.

This is a big refacto even if the API keeps backwards compatibility, so I'd like some input :)

---

This moves the Inference API closer to @huggingface/hub, and conversely, we can also move the @huggingface/hub api closer to the inference module by offering a new HfHub("hf_...") class

[vvmnnnkv]

Really nice to have each task function in its own file!
Some suggestions:

• Put task functions files to a folder like tasks, so they are not mixed with internal methods
• Perhaps even split them to sub-folders like nlp, cv, multimodal, etc. following HF models classification
• Move args/response types to corresponding task function file

[coyotte508]

• Put task functions files to a folder like tasks, so they are not mixed with internal methods

👍

• Perhaps even split them to sub-folders like nlp, cv, multimodal, etc. following HF models classification

And we could move request and streamingRequest to a custom folder too !

• Move args/response types to corresponding task function file

The thing is that some types are reused across tasks. Although we could duplicate the types I guess :) More future-proof

Good suggestions ^^

[coyotte508]

Merging to unblock parallel PRs, waiting a bit before release, feel free to comment in the meantime