docs=705 queries=705
encode seconds=954

 dims bytes/vec    R@1   R@10    MRR
 1024      4096  0.684  0.861  0.752
  768      3072  0.679  0.855  0.749
  512      2048  0.672  0.854  0.740
  256      1024  0.624  0.843  0.705
  128       512  0.586  0.814  0.670
   64       256  0.506  0.779  0.604
1024 int8 bytes/vec=1024  R@1=0.684 R@10=0.861 MRR=0.752

Exact search, 100k random unit vectors, float32, k=10, median of 100 queries, 2 threads
 dims      MB  numpy ms  faiss ms
  256     102      2.51      6.05
  768     307      7.63     16.22
 3072    1229     30.90     69.46

numpy 2.5.4, faiss 1.15.1, sentence-transformers 6.2.0, torch 2.14.1+cpu, python 3.12.15, x86_64