Support BAAI/bge-m3

[hogesako]

Requesting support for BAAI/bge-m3. Thanks.

[NirantK]

[RESOLVED] Blocked on https://github.com/huggingface/transformers/issues/28859

[x4080]

there's already the onnx version, is it compatible with fastembed ?

[x4080]

@NirantK Can I just add "Qdrant/bge-m3-onnx-Q" into models.json (with correct json formatting) ?

[NirantK]

Hey! That model has some loading issues — hence I've not integrated it yet. Somebody from the team should delete that soon from HF as well.

[x4080]

@NirantK I see thanks

[yamalight]

@NirantK is there anything external contributor can do to make this happen? seems like this would be nice first PR 😅

[elsatch]

I see that the block on the transformer library was solved on new releases.

I wonder if part of the loading issues are related to the support from the model of the three kinds of vectors: sparse, dense and multi-vent (colBERT).

[NirantK]

Hey @elsatch - yes, the 2 linear models for Sparse and multi-vec need to have an ONNX export and loader as well. That's something we've to do for now. We can support the dense model without friction but I'd rather avoid the confusion and support all 3 vectors when we support BGE-M3.

[davide445]

Any news on this?

[IneffableBunch]

Any update?

[seetimee]

Any update?

[joein]

Hey, thank you for waiting, we're going to start working on it soon