Binary embeddings support

cozodb / cozo

A transactional, relational-graph-vector database that uses Datalog for query. The hippocampus for AI!

https://cozodb.org

Mozilla Public License 2.0

3.44k stars 108 forks source link

Binary embeddings support #256

Open jimfleming opened 8 months ago

jimfleming commented 8 months ago

Lots of ongoing work using binary embeddings, and other reduced precision embeddings, instead of fp32:

https://txt.cohere.com/int8-binary-embeddings/
https://huggingface.co/blog/embedding-quantization
https://blog.pgvecto.rs/my-binary-vector-search-is-better-than-your-fp32-vectors

Wanted to get thoughts on the appetite for support and if there is appetite what the work might look like in case someone, maybe me, wanted to pick it up. Thanks!

creatorrr commented 7 months ago

yeah, support for binary embeds would be really nice. Correct me if I am wrong but adding naive support would be straightforward right? Could optimize it later on