[GPTQ, CompressedTensors] Fix unsafe imports and metada check

[vasqu]

What does this PR do?

Fixes some unsafe imports for GPTQ and CompressedTensors as well as an unsafe metadata check for GPTQ, i.e. metadata can't be checked if the package doesn't exist...

I tried looking into integrating the env validation function before even creating the quantizers but that would require a complete restructure, especially since the env is checked after creation to get additional info like device map etc. So I kept it as simple as I could and wrapped the unsafe stuff.

Minimal reproducible script to trigger gptq errors (when optimum is not installed):

from transformers import AutoModelForCausalLM, AutoTokenizer, GPTQConfig


model_id = "facebook/opt-125m"
tokenizer = AutoTokenizer.from_pretrained(model_id)
dataset = ["auto-gptq is an easy-to-use model quantization library with user-friendly apis, based on GPTQ algorithm."]
gptq_config = GPTQConfig(bits=4, dataset=dataset, tokenizer=tokenizer)


quantized_model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", quantization_config=gptq_config)

Fixes #34765

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?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@SunMarc @MekkCyber

[MekkCyber]

Thanks for the fix @vasqu, LGTM !

[SunMarc]

Thanks for the fixes !

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[ArthurZucker]

Looks good but I think we should always call validate_environment before even initializing the quantizer (so probably update super (HfQuantizer) to call it at init time no?

[vasqu]

@ArthurZucker
See my comments above:

I tried looking into integrating the env validation function before even creating the quantizers but that would require a complete restructure, especially since the env is checked after creation to get additional info like device map etc. So I kept it as simple as I could and wrapped the unsafe stuff.

The problem is some quantizers need additional information that is not available at creation; another option I could see it to add a post_init to create those unsafe parts :)

[vasqu]

gentle ping @ArthurZucker

[ArthurZucker]

Thanks for explaining

[ArthurZucker]

Not sure if it was fixed by recent quantizer compressed changes, let's resolve conflicts and good to merge

[vasqu]

Forgot about this PR myself 😅 should be good to merge now! Thx for the merges here, ci sometimes do be flaky tho

[MekkCyber]

will merge it once the ci is green @vasqu 🤗