mixedbread-ai

/

mxbai-rerank-xsmall-v1

like 18

@Xenova
Did you test feature-extraction with these weights? It throws the following error for me:

tensor-impl.js:50 Uncaught (in promise) TypeError: dims[1] must be an integer, got: undefined
    at calculateSize (tensor-impl.js:50:19)
    at new Tensor (tensor-impl.js:146:22)
    at new Tensor (tensor.js:78:33)
    at mean_pooling (tensor.js:774:12)
    at Function._call (pipelines.js:1186:84)
    at async embed (worker.js:79:16)
    at async self.onmessage (worker.js:458:25)

Reproducible here with mixedbread-ai/mxbai-rerank-xsmall-v1. The error applies to all onnx mixedbread embedding model versions.

Note that this is not a feature-extraction (embedding) model, it is a text-classification (reranker) model.

For example usage, please refer to the README (JavaScript example).

Oh of course, my bad - mixedbread-ai/mxbai-embed-large-v1 does indeed work as expected!
Thank you!