search

jdagdelen / hyperDB

A hyper-fast local vector database for use with LLM Agents. Now accepting SAFEs at $135M cap.
MIT License
1.38k stars 85 forks source link

Query to understand performance of hyperDB #19

Closed codeastra2 closed 1 year ago

codeastra2 commented 1 year ago

Hi, firstly thanks a lot for this!

I was looking at the implementation of the files and tried to figure out what makes the performance better than other dbs?

  1. I could not find any direct usage of MKL BLAS library , is it somehow internally implemented ?
  2. Is the search proving to be the speed up , with the OpenAI embeddings the metric(cosine similarity) seems to be helping in retrieving the queries faster... It would be great if you could help with my queries thanks!
jdagdelen commented 1 year ago

The numpy backend is hardware accelerated with MKRL and BLAS.