Implement AsyncTextIteratorStreamer for asynchronous streaming

[CISC]

Asynchronous text generation streaming

Added a new AsyncTextIteratorStreamer class that allows you to have fully asynchronous text generation streaming in your application. Works identically to TextIteratorStreamer except that you iterate the stream as follows:

generated_text = ""
async for new_text in streamer:
    generated_text += new_text

Requires Python 3.11+ due to use of asyncio.timeout to handle timeout. Falls back to asyncio.wait_for if asyncio.timeout is not available.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Any hints on how best to implement an async test would be appreciated. :) pytest-asyncio seems to be the way to go...

Who can review?

@gante @ArthurZucker

[CISC]

Hmmm, looks like I'm missing something important regarding documentation, not sure what though?

Exception: The following objects are in the public init so should be documented:

• AsyncTextIteratorStreamer

Was missing autodoc entry.

[CISC]

@gante @ArthurZucker Ready for review.

The check_code_quality workflow failure makes no sense to me, not sure how to fix?

[CISC]

@ArthurZucker @gante ping

[CISC]

@ArthurZucker ping with sugar on top?

Seems @gante is on holiday, anyone else who can review?

[Rocketknight1]

Gentle ping @gante, or maybe @zucchini-nlp if I haven't totally overloaded her with pings at this point!

[CISC]

@Rocketknight1 Thanks! BTW, know what's up with check_code_quality and/or how to fix it?

[Rocketknight1]

@CISC that's usually a code style thing. Try pip install transformers[quality] followed by make fixup

[CISC]

@Rocketknight1 It's not, it's complaining about the import order in src/transformers/__init__.py, which I did not change... :(

[Rocketknight1]

@CISC make fixup will also resolve issues like that! In this case, the __init__.py has actually been modified in this PR to allow the class to be imported (check Files changed - it's in there!). Our code styling tools just want to correctly sort/order inputs into the standard format.

[CISC]

@Rocketknight1 Sure, I changed the file, but I didn't change the sort order, it should already be sorted! :P

[ArthurZucker]

Sorry @CISC reviewing!

[ArthurZucker]

Looks marvelous sorry for being so late. I actually wanted to wait because we added huggingface/tokenizers#1678 a decode stream api in tokenizers that fixes some issues with prefix space, but the changes are for another PR and for the base Text streamer! 🤗 cc @gante

[ArthurZucker]

Do you want us to fix the CIs for you? I can push directlyu

[CISC]

Do you want us to fix the CIs for you? I can push directlyu

Yes please, thank you. :)

[ArthurZucker]

thanks @CISC 🤗