-
Notifications
You must be signed in to change notification settings - Fork 25.6k
Open
Labels
:Search Relevance/VectorsVector searchVector search>enhancementTeam:Search RelevanceMeta label for the Search Relevance team in ElasticsearchMeta label for the Search Relevance team in Elasticsearch
Description
Description
We block encode centroids into blocks of 16, which is fine for when they don't also have their own parents, but continually, we end up having clusters of centroids that are less than 16, which harms query throughput significantly.
Consequently, we should always block encode centroids, even a "tail" that is less than 16 vectors in size.
In parallel, we likely should increase the block size. Bulk scoring off heap provides a significant speed up for centroid scoring (see: #138204)
Metadata
Metadata
Assignees
Labels
:Search Relevance/VectorsVector searchVector search>enhancementTeam:Search RelevanceMeta label for the Search Relevance team in ElasticsearchMeta label for the Search Relevance team in Elasticsearch