oneshot entrypoint update

[ved1beta]

SUMMARY:
Updated the oneshot function to use explicit keyword arguments instead of relying solely on **kwargs.

TEST PLAN:
Changes were tested using mock tests that verify:
Correct handling of explicit arguments
Backward compatibility with kwargs
Proper behavior with mixed explicit args and kwargs
Verification that kwargs correctly override explicit parameters when duplicates exist
fixes
#1443

[github-actions]

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

Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed.

[kylesayrs]

This looks great! There are a few arguments missing from the signature like dvc_data_repository, but this is handled and by **kwargs and those arguments don't need to be explicit.

This will go a long way towards making LLM Compressor more user friendly :)

[kylesayrs]

In the future we could use something like this to avoid duplicating information

from typing import Tuple, List, Any, Dict
import inspect

def collect_function_arguments() -> Tuple[List[Any], Dict[str, Any]]:
    caller_frame = inspect.currentframe().f_back
    args_info = inspect.getargvalues(caller_frame)
    args = []
    kwargs = {}
    for key, value in args_info.locals.items():
        if key in args_info.args:
            args.append(value)
        elif key == args_info.varargs:
            args.extend(value)
        elif key == args_info.keywords:
            kwargs.update(value)
        else:
            assert False
        
    return args, kwargs
    
def oneshot(...):
    args, kwargs = collect_function_arguments()
    one_shot = Oneshot(*args, **kwargs)

[kylesayrs]

@ved1beta Could you please make style; make quality?

[brian-dellabetta]

thanks @ved1beta for taking a shot at this! I think some of the type hints need to be updated to account for actual classes being passed in other than string representations, but otherwise this is looking great!

[brian-dellabetta]

Users can also pass in a pre-loaded model

Suggested change

	model: str,
	model: Union[str, PreTrainedModel],

[brian-dellabetta]

These can also be the loaded up classes. I think these are the best base classes to use? cc @kylesayrs

Suggested change

	tokenizer: Optional[str] = None,
	processor: Optional[str] = None,
	tokenizer: Optional[Union[str, transformers.PreTrainedTokenizerBase]] = None,
	processor: Optional[Union[str, transformers.ProcessorMixin]] = None,

[brian-dellabetta]

This can also be a list of Modifiers

[brian-dellabetta]

this can be an actual loaded dataset