[ 
https://issues.apache.org/jira/browse/SOLR-18267?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18130369#comment-18130369
 ] 

ASF subversion and git services commented on SOLR-18267:
--------------------------------------------------------

Commit 66e43fdfe0de0fc563c41101735e67d28442ca3e in solr's branch 
refs/heads/branch_9x from Adam Quigley
[ https://gitbox.apache.org/repos/asf?p=solr.git;h=66e43fdfe0d ]

SOLR-18267: Add flat vector index with no HNSW (#4492)

Adapted for branch_9x: BinaryQuantizedDenseVectorField (not on 9.x) and the
cagra_hnsw algorithm references were dropped; tests and docs adjusted to 9.x.

(cherry picked from commit 07036ab1c700a1ead8282482d495b17b17bf0387)
(cherry picked from commit bec1acb9af50f5969f570349d3f535359001ab18)


> Skip HSNW graph construction for exact vector search
> ----------------------------------------------------
>
>                 Key: SOLR-18267
>                 URL: https://issues.apache.org/jira/browse/SOLR-18267
>             Project: Solr
>          Issue Type: New Feature
>          Components: vector-search
>            Reporter: Alessandro Benedetti
>            Assignee: Alessandro Benedetti
>            Priority: Minor
>              Labels: pull-request-available, vector-based-search
>             Fix For: 10.1
>
>          Time Spent: 5h 10m
>  Remaining Estimate: 0h
>
> The idea we are exploring is not building the HNSW graph in the case of a 
> highly selective filter search use case, and performing brute force KNN 
> comparison as a post-filter.
> The goal behind the feature is to not pay the cost of HNSW building during 
> ingestion when ANN search isn't used.
>  
> A collaborator of ours already created a proof-of-concept on a branch of 
> upstream Solr that adds knnAlgorithm="flat" to DenseVectorField using 
> Lucene99FlatVectorsFormat. This stores vectors in the index (.vec/.vemf 
> files) without building the HNSW graph (.vex/.vem files).
>  



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to