added model_kwargs to huggingface model

[nanbo-liu]

This is for issue 1344
This allow model-settings.json to take model_kwargs:
example below:

{
    "name": "llama",
    "implementation": "mlserver_huggingface.HuggingFaceRuntime",
    "parameters": {
        "extra": {
            "task": "text-generation",
            "pretrained_model": "daryl149/llama-2-7b-hf",
            "model_kwargs": {
                "device_map": "auto",
                "load_in_8bit": true
            }
        }
    }
}

[nanbo-liu]

@adriangonz ,could you review it when you have a chance

[adriangonz]

Nice one @nanbo-liu! This would be a great addition to the HF runtime - thanks for contributing it 🚀

Changes look good - although, could you add some tests to validate that the model kwargs get used? Once tests are in, we should be good to land this one. 👍

[nanbo-liu]

@adriangonz ,added unit tests from the latest commit.

[adriangonz]

Brilliant! Thanks for the changes @nanbo-liu !

This should be good to go now 🚀

[nanbo-liu]

@adriangonz ,I rewrite the unit test a little bit. The previous version depends on package accelerate.
I don't want to introduce extra dependency into this repo. Would you take another look when you have a chance.
Need to rerun workflow

[nanbo-liu]

@adriangonz, just fixed the unit test bug. would you try again

[nanbo-liu]

@adriangonz ,CICD keeps getting runtime error for catboost.
I see other PRs also got runtime error for catboost

[adriangonz]

Yuuup, it seems there's an unrelated error in the catboost runtime causing some actions to fail. Which means this PR should be good to go as-is 👍