cross-encoder not improving accuracy in semantic search

[gypsydave5]

Hey! I'm coding along with the course and I thought I might have made some mistake - the semantic search results re-ranked with the cross encoder have the same accuracy as the results before re-ranking.

So I pulled the notebook from this repo to run it... and it's also the same.

Has one of the models changed under the hood in the last six months? Is there anything I could try out to see a cross-encoder at work?

(enjoying the book + course otherwise - ty!)

[sinanuozdemir]

Hey David, thanks so much for letting me know! Just to be certain, you're talking about numbers from 02_semantic_search.ipynb, right? It's showing that accuracy went up after re-ranking but I'll have to re-run the code myself when I get the chance to see what may have gone wrong here!

In the meantime, here's another example of re-ranking from a RAG course I give here. I hope this helps! It's using a closed-source re-ranker though just a heads up, not open-source.

I really do appreciate you bringing this to my attention!