[v2] Updated ModelMeta parameters

[Samoed]

Currently, the model annotation looks like this:

e5_mult_small = ModelMeta(
    loader=partial(  # type: ignore
        sentence_transformers_loader,
        model_name="intfloat/multilingual-e5-small",
        revision="fd1525a9fd15316a2d503bf26ab031a61d056e98",
        model_prompts=model_prompts,
    ),
    name="intfloat/multilingual-e5-small",
    revision="fd1525a9fd15316a2d503bf26ab031a61d056e98",
    ...
)

This causes code duplication with the model name and revision being repeated. I propose changing this to pass the model name and revision directly to the loader using get_model, as demonstrated [here](

mteb/mteb/model_meta.py

Lines 120 to 125 in 0c5c3a5

	loader = partial(
	sentence_transformers_loader,
	model_name=self.name,
	revision=self.revision,
	**kwargs,
	)

).

With this approach, ModelMeta could look like this:

e5_mult_small = ModelMeta(
    loader=sentence_transformers_loader,  # callable or class
    loader_kwargs={"model_prompts": model_prompts},  # dict with additional kwargs
    name="intfloat/multilingual-e5-small",
    revision="fd1525a9fd15316a2d503bf26ab031a61d056e98",
    ...
)
# model_meta.py
model = meta.loader(meta.name, meta.revision, similarity_fn_name=meta.similarity, **meta.loader_kwargs, **kwargs) 

This would reduce code duplication and make the implementation cleaner.

Issue discussion start #1759

[KennethEnevoldsen]

Thanks for outlining it. I am perfectly fine with this case