jimczi commented on PR #16769: URL: https://github.com/apache/lucene/pull/16769#issuecomment-6035960746
Thanks @ChrisHegarty! I've narrowed this since your review, to fixing #16749's merge mapping and covering Faiss: - The de-duplicating formats are out. Their merges read the raw vectors mostly forward with jumps back for duplicates, close to how searches read them, and their raw file mixes vectors the graph walks with vectors kept only to rescore. Doing them right needs a format change, so I'll take that separately. - The merge mapping no longer decides on reuse: it keeps the caller's hints and only switches the access to sequential. No-reuse during merges now applies where the format says so, like the scalar quantized format's raw vectors, and not to HNSW's raw vectors. -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
