Closed metonymic-smokey closed 4 months ago
@metonymic-smokey did you have a document of your testing with SQ4? I'm looking specifically for if this memory usage improvement translates with large number of vectors within a single index - say 100000+.
@metonymic-smokey did you have a document of your testing with SQ4? I'm looking specifically for if this memory usage improvement translates with large number of vectors within a single index - say 100000+.
@abhinavdangeti posted this on the linked ticket.
thanks!
Thanks @metonymic-smokey 👍🏼
Rebased, and bring in https://github.com/blevesearch/bleve_index_api/pull/49 .
This PR changes the quantiser to SQ4 from SQ8 when the index is optimised for memory.