Postgress

The score is purely based on however postgres calculates it 👀

Some embedding models give different scores. A "good" score for one model might be very different compared to another model. It's all pretty relative tbh

Just need to make sure you query the index with the same embedding model that you used to construct it

@Logan M That might be it! Is there a way for me to modify the scoring behavior?

Thanks for taking the time to answer 🙂

hmm, with postgres, I'm actually not sure 😅

Thanks for the response Logan, I always appreciate it 🙂 @Logan M

I do my best :LlamaIndex:

Quick thought - do you think re-ranking the nodes will help me display at least one source? I'm using the same embedding model as in the query engine and the answers are SPOT ON, but the score is 0.45 or so and the 0.5 threshold has become a product requirement

(imo seems a little silly to set a arbitrary threshold like that, it should be specific to the data and embed model)

But yea, reranking will change the scores, so it could help

Thanks 🙂