[Oneshot] Oneshot Refactor

[horheynm]

SUMMARY:
Edit oneshot pathway to use only the necessary code.

Problem:
Oneshot uses a pathway that is shared in other entrypoints, eg. train.
Oneshot doesnt train, so do not need use code that trains, as an example of redundant code. Remove other parts of the code and refactor

Design

Changes:

• Oneshot class responsible for carrying out the oneshot pipeline
• Oneshot run will not depend on the session. Only use CompressionLifecycle that has modifer logic support. Places that uses active_session, such as saving to the compressed model by reading the stagemodifiers, has been replaced by passing in the necessary args

TEST PLAN:
Pass existing passing tests
Pass all finetune tests merging transformer main for HFQuantizer support
Verified new pathway has the same scales for the int8 w8a8 dynamic per token values for Meta-Llama-3-8B-Instruct

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

[dsikka]

It seems like we're introducing more abstractions, not fewer, when separating out the oneshot entrpoint and I'm not sure these are necessary.

we also seem to be repeating a lot of code everywhere.

[dsikka]

SUMMARY: Edit oneshot pathway to use only the necessary code.

Problem: Oneshot uses a pathway that is shared in other entrypoints, eg. train. Oneshot doesnt train, so do not need use code that trains, as an example of redundant code. Remove other parts of the code and refactor

Changes:

• Oneshot run will not depend on the session. Only use CompressionLifecycle that has modifer logic support.
• Return model as a return type to oneshot to obtain model without relying on session.

TEST PLAN: Pass existing passing tests Verified new pathway has the same scales for the int8 w8a8 dynamic per token values for Meta-Llama-3-8B-Instruct

Please update the PR desription showing what the updated entrypoint looks like as a result of this PR.

[horheynm]

@dsikka Thanks for looking at the pr early. Yes thats the next to do before the full review. One ready i will ping you!

[horheynm]

Only used in HF training args
https://github.com/huggingface/transformers/blob/main/src/transformers/training_args.py#L819