benwtrent commented on PR #15564:
URL: https://github.com/apache/lucene/pull/15564#issuecomment-3745127557
Here is top-100 (hack, bit=3 means 2 bits and 8 bit queries, I did that so I
could quickly test them together):
1M 768 coherev2 max inner product.
```
recall latency(ms) netCPU avgCpuCount fanout quantized visited
index(s) index_docs/s force_merge(s) index_size(MB)
0.754 0.953 0.931 0.977 0 2 bits 5109
293.10 3411.82 250.63 3221.63
0.776 1.669 1.649 0.988 100 2 bits 8853
293.10 3411.82 250.63 3221.63
0.780 2.050 2.028 0.989 150 2 bits 10572
293.10 3411.82 250.63 3221.63
0.783 2.443 2.418 0.990 200 2 bits 12202
293.10 3411.82 250.63 3221.63
0.761 1.007 0.993 0.986 0 3 bits 5113
291.79 3427.10 167.24 3222.90
0.783 1.765 1.758 0.996 100 3 bits 8881
291.79 3427.10 167.24 3222.90
0.787 2.238 2.230 0.996 150 3 bits 10622
291.79 3427.10 167.24 3222.90
0.790 2.740 2.656 0.969 200 3 bits 12269
291.79 3427.10 167.24 3222.90
```
Trying 1024dim cohere now.
--
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]
