Do off-heap scoring for int8_flat quantized vectors

elastic / elasticsearch

Free and Open, Distributed, RESTful Search Engine

https://www.elastic.co/products/elasticsearch

Other

68.51k stars 24.33k forks source link

Open benwtrent opened 4 weeks ago

benwtrent commented 4 weeks ago

In newer versions of Elasticsearch, we have an optimized ARM implementation for int8_hnsw vectors.

However, we are not using the the off-heap optimized scorer for int8_flat, we should try to do this if possible.

This is only applicable after our upgrade to Lucene 9.11

elasticsearchmachine commented 4 weeks ago

Pinging @elastic/es-search (Team:Search)