LlamaIndex
Log in

Find answers from the community

Updated 2 years ago

Pgvector

At a glance

The post indicates that the community member has moved to PostgreSQL. The comments discuss the use of the pgvector library and the Supabase vector store, and whether they are returning distance or similarity. Some community members are unsure about the differences between the PostgreSQL adaptor and the Supabase version, but they conclude that the differences are not significant.

Useful resources
SSeaCat
·
Yes, I moved to postgresql.
L
S
14 comments
LLogan M
Hmm, the code just uses the raw similarity returned by pgvector.

I haven't looked into it much, but maybe pgvector is returning distance instead of similarity? So smaller is closer?
LLogan M
https://github.com/jerryjliu/llama_index/blob/76f3310b3f887bfb48f32cefcb5d5601b5805bb8/llama_index/vector_stores/postgres.py#L201
LLogan M
Should probably find pgvector docs for this
SSeaCat
Wait a sec. I didn't know you have your own postgres adaptor. I actually used the supabase version
LLogan M
Oh! Yea we support both pgvector/postgres and supabase/vecs
LLogan M
Supabase similarity is here
https://github.com/jerryjliu/llama_index/blob/76f3310b3f887bfb48f32cefcb5d5601b5805bb8/llama_index/vector_stores/supabase.py#L166
LLogan M
Looks like it's assuming it's distance 🤔
SSeaCat
what means the less the better, right?
LLogan M
Yea, that's what distance means. And we are converting the distance to a similarity

But it seems like it probably isn't returning distance
SSeaCat
Oh I see it's in the query function. Thanks!!
SSeaCat
btw what is the difference between your postgres adaptor and supabase's one?
LLogan M
Tbh, I really have no clue haha

I think they both use postgres, but supabase uses this vecs library instead of pgvector? Maybe?
SSeaCat
they use pgvector on top of postgres
SSeaCat
yeah I don't think the difference is significant
Add a reply
Sign up and join the conversation on Discord