Open question: HNSW vs exhaustive KNN when number of chunks is around 10k

[cherifbenham]

Hello friends
Does somebody know if there is a significant gain in accuracy/recall when using exhaustive knn over hnws?
I guess the search latency/speed is still reasonable for n_chunks around 10k
What are your thoughts on this?

Thanks !!!!

[ross-p-smith]

I would say that this depends on the data/index you are using. I would look at something like the https://github.com/microsoft/rag-experiment-accelerator to look at the gains and when you have finished then configure this repository based on your answer