consider using vek

[josharian]

I ran a very quick benchmark, and https://github.com/viterin/vek appears to be ~10x faster for cosine similarity than plain Go on an AVX2 machine.

Given this from the README:

At 1536 dimensions (OpenAI default), 70% of the query process under default parameters is spent in the distance function.

It seems like using vek could have a pretty big impact.

[ammario]

c8a7f6c