Lajavaness

/

bilingual-embedding-base

like 2

The model seem to report the variant of the model as 'bilingual' instead of 'xlm-roberta', which causes text-embeddings-inference (https://github.com/huggingface/text-embeddings-inference) to refuse to use the model.
Is this motivated by an architecture change ?

To reproduce

model=Lajavaness/bilingual-embedding-base
volume=$PWD/data

docker run --platform linux/x86_64 -p 8080:80 -v $volume:/data --pull always ghcr.io/huggingface/text-embeddings-inference:cpu-1.4 --model-id $model

Output

2024-07-05T09:55:48.701830Z  INFO text_embeddings_router: router/src/lib.rs:250: Starting model backend
Error: Could not create backend

Caused by:
    Could not start backend: Model is not supported

    Caused by:
        unknown variant `bilingual`, expected one of `bert`, `xlm-roberta`, `camembert`, `roberta`, `distilbert`, `nomic_bert`, `mistral`, `new`, `qwen2` at line 6 column 27