Be careful: they have super short context length AND silently crop if the text is too long. To me there is really no reason to use them.

I recommend ollama to run the artic-embed-v2 model, it also is multimingual and you can use --quantize when loading the modelfile to get it even smaller.