vigyasharma commented on PR #13525: URL: https://github.com/apache/lucene/pull/13525#issuecomment-2911279145
> if our goal were limited to enabling MAXSIM scoring between two multi-vectors using only exact search (i.e., without KNN indexing for faster retrieval), many of these challenges could be avoided. one of the challenges, and perhaps the main concern with this PR, has been practical scaling challenges of working with multi-vectors on the entire corpus. This would apply to exact search as well. A viable middle ground would be to over-collect documents via regular single-vector knn search in the first pass, and later rerank them using late-interaction multi-vectors in the second pass. This idea was shared by jimczi@ earlier in [this thread](https://github.com/apache/lucene/pull/13525#issuecomment-2445295372). I'm working on a PR that combines his approach with a FunctionScore query for reranking. Should be able to raise it for review this week. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
