iprithv commented on PR #15982: URL: https://github.com/apache/lucene/pull/15982#issuecomment-4509725559
> Thanks for running the tests. Change wise it looks good to me. > > > for float vectors, nothing changes. > > This fix appears to have no impact to float vectors which you earlier asserted as well. Because we are not touching/changing anything around float vectors accounting (right?). But from your results above why does the no. of segments seems to be materially different/reducing for float32 case(9 -> 3)?. @shubhamvishu I re-ran the float32 benchmark on main and got 3 segments this time (same as fix). looks like the earlier 9 segments was just run-to-run noise, merge timing can vary a bit. for float32, nothing really changes here: - BufferingKnnVectorsWriter change is basically a no-op (still 4 bytes) - benchmark is using Lucene99HnswVectorsFormat, so quantized writer changes don’t apply so float32 behavior is unchanged. the small time differences (~1–2%) and segment count changes are just noise. the only real change is for byte vectors in quantized formats, where they are now they are tracked correctly -- 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]
