Low cosine similarity score when using Langchain FAISS wrapper and CLIP Model

pasta_paul

I am currently using Langchain wrapper of FAISS vector store and CLIP model. Since Langchain wrapper of FAISS supports only text embedding models , I have implemented my own embedding class(using multimodal embedding - CLIP) that implements the abstract methods of abstract class. For testing, I gave a few Images as an input and stored them as embedding in FAISS. Then when I passed a text query, I am getting an accurate match with image but the...

Originally asked by Aashishkumar on Stack Exchange (stackoverflow) (CC BY-SA 4.0).

uphilltech

Good thread on this — worth a read before you pay anyone to build it.

Discussion: https://stackoverflow.com/questions/79921783/low-cosine-similarity-score-when-using-langchain-faiss-wrapper-and-clip-model.

Want it set up for you? UphillTech.