Tejas Shah
Tejas Shah
It defaults to index ef_ value if its not type HNSWQuery This introduces the ability to pass in ef value at query time. https://github.com/nmslib/nmslib/issues/442
### Description efSearch is one of the parameters in HNSW algorithm which can help increase recall. efSearch value indicates the neighbors that the algorithm explores to get the nearest ones...
### Description This approach extends Lucene's KNNFloatVectorQuery. Testing in progress There are some differences and its hard to achieve status quo with current implementation without overriding rewrite or making changes...
[DO NOT MERGE][NEED INITIAL FEEDBACK]Introduces NativeEngineKNNQuery which executes ANN on rewrite
### Description Pending unit test, All integration tests are passing. Cannot support radial currently since k is not significant, there are ways to do it but since its not planned...
**Is your feature request related to a problem?** Currently we are using traditional pointer and deleting it in case of errors in nmslib_wrapper. This is done to be able to...
**Is your feature request related to a problem?** Currently adding parameters in K-NN requires changes in Neural. There is a chance of inconsistencies there with this duplication and increases the...
### Description Currently live docs are computed with linear complexity even when it does not have deleted docs. This PR skips computing it #### Testing - UTs and javadocs in...
**What is the bug?** After the switch to `NativeEngines990KnnVectorsWriter` we saw force merge latencies increased approximately by [20%](https://benchmarks.opensearch.org/app/dashboards#/view/f9ef2c40-ee11-11ee-a589-033f2c21734f?_g=(filters:!(),refreshInterval:(pause:!t,value:0),time:(from:now-15m,to:now))&_a=(description:'',filters:!(('$state':(store:appState),meta:(alias:!n,disabled:!f,index:'333e2950-edec-11ed-a453-51f8c622cf9a',key:query,negate:!f,type:custom,value:'%7B%22bool%22:%7B%22minimum_should_match%22:1,%22should%22:%5B%7B%22match_phrase%22:%7B%22user-tags.cluster-config%22:%22arm64-r6g.4xlarge-3-data-6-shards-1-replica-nmslib-cohere-10m%22%7D%7D,%7B%22match_phrase%22:%7B%22meta.tag_cluster-config%22:%22arm64-r6g.4xlarge-3-data-6-shards-1-replica-nmslib-cohere-10m%22%7D%7D%5D%7D%7D'),query:(bool:(minimum_should_match:1,should:!((match_phrase:(user-tags.cluster-config:arm64-r6g.4xlarge-3-data-6-shards-1-replica-nmslib-cohere-10m)),(match_phrase:(meta.tag_cluster-config:arm64-r6g.4xlarge-3-data-6-shards-1-replica-nmslib-cohere-10m))))))),fullScreenMode:!f,options:(hidePanelTitles:!f,useMargins:!t),query:(language:kuery,query:''),timeRestore:!f,title:'%5BVector%20Search%5D%20%5BNmslib%5D%201OM%20Cohere%20768D',viewMode:view)) in nightly runs The increase has been consistent **How can one reproduce...
## Description Exact search evaluates vectors in linear fashion. Leveraging IndexInput#prefetch to load the next vector in memory, can possibly help with reducing the read cost during runtime reducing the...
### Description Adds exact and ann timings and counts with knn query profile API You can find sample response here: [concurrent_nofilter.json](https://github.com/user-attachments/files/18564364/concurrent_nofilter.json) [concurrent.json](https://github.com/user-attachments/files/18564356/concurrent.json) [profile_filter.json](https://github.com/user-attachments/files/18564357/profile_filter.json) [faiss_profile.json](https://github.com/user-attachments/files/18564358/faiss_profile.json) CIs will fail since it require...