benwtrent commented on issue #14007: URL: https://github.com/apache/lucene/issues/14007#issuecomment-2491059836
> The technical feasibility of this new storage model. Very feasible. However, it gets tricky on accuracy and the quantization methods will require updating to handle this. But all the current (and hopefully future) quantization methodologies can easily "re-hydrate" vectors so that iterating floats is still possible (folks give floats, we should be able to return floats if they want them). > Potential impacts on search accuracy and performance. For higher fidelity quantization methods, I expect accuracy to be OK. I am not 100% sure with how things are now if we would ever want to quantize below int7 for storage. Int4 as it stands (when using the dynamic confidence interval), would suffer greatly without some more significant improvement to the algorithm. -- 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]
