Why the success rate of hitTesting deployed through jinaai/jina-embeddings-v2-base-zh privatization is far lower than that of hitTesting called by jina's api

#10
by edisonzf2020 - opened

I accessed jina api and privately deployed jinaai/jina-embeddings-v2-base-zh in the dify system, and used them as embedding models to test the same document. The jina api effect was very good, but the private deployment was completely unqualified. .

Sign up or log in to comment