undefined | Better HN

0 pointsgsuuon2y ago0 comments

So to rephrase: stage 1 - it finds the top k most relevant small chunks, stage 2 - it searches the source documents of those small chunks for most relevant large chunks, stage 3 - it searches medium chunks in the source documents of relevant small chunks that are contained in the large chunks found in stage 2?

0 comments

1 comments · 1 top-level

joeyxiong2y ago

Yeah, the sliding window chunking mechanism appears in stage 2.

j / k navigate · click thread line to collapse