feat: Add .from_hf_hub() and similar methods to ModelMeta

[Samoed]

Close #3735
Close #3695
Close #3734

Move _model_meta_from_hf_hub, _model_meta_from_cross_encoder, _model_meta_from_sentence_transformer to ModelMeta class

from sentence_transformers import SentenceTransformer

from mteb.models import ModelMeta

model = SentenceTransformer("Qwen/Qwen3-Embedding-0.6B", device="cpu")
meta = ModelMeta.from_sentence_transformer_model(model)
print(meta.to_dict())
# {'loader_kwargs': {}, 'name': 'Qwen/Qwen3-Embedding-0.6B', 'revision': 'c54f2e6e80b2d7b7de06f51cec4959f6b3e03418', 'release_date': None, 'languages': None, 'n_parameters': 595776512, 'memory_usage_mb': 1136, 'max_tokens': 32768, 'embed_dim': 1024, 'license': 'apache-2.0', 'open_weights': True, 'public_training_code': None, 'public_training_data': None, 'framework': ['Sentence Transformers'], 'reference': None, 'similarity_fn_name': <ScoringFunction.COSINE: 'cosine'>, 'use_instructions': None, 'training_datasets': None, 'adapted_from': None, 'superseded_by': None, 'modalities': ['text'], 'is_cross_encoder': None, 'citation': None, 'contacts': None, 'loader': 'sentence_transformers_loader'}

[KennethEnevoldsen]

We can fetch similarity function from: https://huggingface.co/sentence-transformers/embeddinggemma-300m-medical/blob/main/config_sentence_transformers.json#L24

but if not I would be ok with assuming cosine? (I suspect it is also in the model data).

Feel free to make this a seperate issue

[KennethEnevoldsen]

might be easier to just fetch from the model - but if we could avoid loading/downloading the model that would be great

[Samoed]

Added from_hub_for_sentence_transformer maybe not best naming

[KennethEnevoldsen]

Hmm I don't think that is what we want. Why can't that just be the default for from_hub?

[Samoed]

Renamed

[KennethEnevoldsen]

Suggested change

	meta = cls.from_hf_hub(model.model.name_or_path, revision, compute_metadata)
	meta = cls.from_hf_hub(model.model.name_or_path, revision, compute_metadata)

maybe worth splitting this into an inner _fetch_metadata_from_hub
and a public fetch_from_hub.

[KennethEnevoldsen]

1. Let us add a warning here
2. Shouldn't this be based on the number of parameters? It could have API tag while also having public weights

[Samoed]

Currently, we have api tag only for private models, so I'm not sure what to do here

[KennethEnevoldsen]

But if it is private, then the number of parameters is None? (We have a few private functions where the number of parameters is public, but then I suppose we could also estimate the memory usage?). Though I see that might be a bit odd

[Samoed]

Added to the doc

[Samoed]

FYI This would also close #3695 #3734 (added to devolompent section)

[Samoed]

Problem with HF limits again or HF have some problems, but there is nothing in their status page (aws seems fine too)

[KennethEnevoldsen]

Hmm I don't think that is what we want. Why can't that just be the default for from_hub?

[KennethEnevoldsen]

I think we are at a solid state!