Re: [I] Increase the number of dims for KNN vectors to 2048 [LUCENE-10471] [lucene]

["jzwolak (via GitHub)" <gi...@apache.org>]

jzwolak commented on issue #11507:
URL: https://github.com/apache/lucene/issues/11507#issuecomment-2640596317

   @asfimport
   
   I suggest making it easier to change the limit. I appreciate the need to have a limit for optimization and performance. The models that have more than 1024 dimensions in their embedding vectors are not bad models or poor choices. In fact, there are models that have millions of dimensions. And there's real value there. Though I will add that the model I heard about at a conference talk with millions of dimensions did simplify the vector to have base-2 integers instead of floating point numbers. That isn't compatible with what Lucene is doing here, but my point remains that there are larger models and valid reasoning for such models.
   
   I'm working with a model with 3072 dimensions and getting _great_ results. Reducing the dimensions will lose information.
   
   Last, other vector search indexers/databases support very quick search for 3072 dimensions. Like Qdrant.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org