High scores

#8
by drmeir - opened

The model always produces high similarity scores. For example, the similarity score between the word King and the word Dog is 0.796. I was not able to get a score below 0.7 for any pair of words. This seems wrong... What am I missing? How do I get scores that make intuitive sense?

Please refer to the related discussions at https://huggingface.co/intfloat/multilingual-e5-large/discussions/10 and https://github.com/microsoft/unilm/issues/1216

I will update the model card to add some clarification to avoid future confusions.

Sign up or log in to comment