benwtrent commented on PR #15903:
URL: https://github.com/apache/lucene/pull/15903#issuecomment-4180477312
I ran the 400k cohere v3 set on my machine, TQ does seem to provide nicer
quantization mechanics on low bits. However, its really tricky to get
performance right with LUTs for all CPUs in Java land :/
```
recall latency(ms) netCPU avgCpuCount quantized visited index(s)
index_docs/s force_merge(s) index_size(MB) vec_disk(MB) vec_RAM(MB)
0.651 1.630 1.624 0.997 1 bits 9683 121.66
3287.88 62.24 1679.17 1616.669 54.169
0.772 2.051 2.037 0.993 2 bits 8591 126.26
3168.17 80.05 1722.58 1665.497 102.997
0.905 2.442 2.424 0.993 4 bits 8061 149.44
2676.59 93.41 1820.22 1763.916 201.416
0.979 3.448 3.431 0.995 8 bits 7960 187.99
2127.78 220.49 2015.69 1959.229 396.729
0.989 2.303 2.269 0.985 no 7952 197.75
2022.74 ```
--
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]
