kaivalnp commented on PR #15802: URL: https://github.com/apache/lucene/pull/15802#issuecomment-4014899955
`knnPerfTest.py` benchmarks (using https://github.com/mikemccand/luceneutil/pull/542) with Cohere v3, 1M docs, 10K queries: Baseline with `-numSearchThread=2`: ``` recall latency(ms) netCPU avgCpuCount nDoc searchType topK fanout traversalSimilarity resultSimilarity maxConn beamWidth quantized visited num_segments filterStrategy filterSelectivity overSample bp-reorder indexType 0.956 0.766 1.529 1.996 1000000 RADIUS N/A N/A 0.78 0.8 64 250 8 bits 2912 5 null N/A 1.000 false HNSW ``` Candidate with `-numSearchThread=2`: ``` recall latency(ms) netCPU avgCpuCount nDoc searchType topK fanout traversalSimilarity resultSimilarity maxConn beamWidth quantized visited num_segments filterStrategy filterSelectivity overSample bp-reorder indexType 0.956 0.734 1.569 2.139 1000000 RADIUS N/A N/A 0.78 0.8 64 250 8 bits 2912 5 null N/A 1.000 false HNSW ``` Baseline with `-numSearchThread=8`: ``` recall latency(ms) netCPU avgCpuCount nDoc searchType topK fanout traversalSimilarity resultSimilarity maxConn beamWidth quantized visited num_segments filterStrategy filterSelectivity overSample bp-reorder indexType 0.956 0.757 1.515 2.001 1000000 RADIUS N/A N/A 0.78 0.8 64 250 8 bits 2912 5 null N/A 1.000 false HNSW ``` Candidate with `-numSearchThread=8`: ``` recall latency(ms) netCPU avgCpuCount nDoc searchType topK fanout traversalSimilarity resultSimilarity maxConn beamWidth quantized visited num_segments filterStrategy filterSelectivity overSample bp-reorder indexType 0.956 0.682 1.595 2.337 1000000 RADIUS N/A N/A 0.78 0.8 64 250 8 bits 2912 5 null N/A 1.000 false HNSW ``` -- 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]
