[New Model]: BAAI/bge-m3

[Yikun]

The model to consider.

This issue tracks initial support for the embedding model with vllm-ascend:

https://huggingface.co/BAAI/bge-m3

The closest model vllm already supports.

No response

What's your difficulty of supporting the model you want?

No response

[Yikun]

P0:

Model	Arch	type
https://huggingface.co/BAAI/bge-m3	XLMRobertaModel	embed
https://huggingface.co/BAAI/bge-reranker-v2-m3	XLMRobertaForSequenceClassification	score
https://huggingface.co/BAAI/bge-large-zh-v1.5	BertModel	embed

P1:

Model	Arch
https://huggingface.co/BAAI/bge-reranker-large	XLMRobertaForSequenceClassification
https://huggingface.co/BAAI/bge-large-en-v1.5	BertModel
https://huggingface.co/BAAI/bge-base-en-v1.5	BertModel
https://huggingface.co/BAAI/bge-small-en-v1.5	BertModel
https://huggingface.co/BAAI/bge-reranker-base	XLMRobertaForSequenceClassification
https://huggingface.co/BAAI/bge-large-zh	BertModel
https://huggingface.co/jinaai/jina-embeddings-v3	XLMRobertaForSequenceClassification

[MengqingCao]

FYI
I'm working on BAAI/bge-reranker-v2-m3, which requires encoder-only attention. Now we're facing a precision issue with it, will update the progress here in time.