msokolov commented on issue #13251:
URL: https://github.com/apache/lucene/issues/13251#issuecomment-4327164110

   If we were to a dd a config option what would it look like?  I guess the key 
elements are: quantization is one-and-done; no re-evaluation on merging, and 
therefore any data required only for indexing can be dropped.  This could be 
true whether or not some higher-precision vectors are preserved for re-ranking. 
I mean it's conceivable we'd want to quantize at 1 bit for recall and 8 bits 
for re-ranking?  To me the key choice is whether to re-evaluate quantization 
bounds on merge.  I actually wonder if with OSQ this is ever necessary since it 
is computing bounds per-vector?


-- 
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: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


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

Reply via email to