Currently this is a proof of concept Postgresql plugin for text embeddings.
Goal is to use Flame with PGVector for faster RAG applications in Postgresql.
Model chosen is MiniLM-L6-v2
, since I was using it already with 768 length.
The proof of concept (v0.0.1) uses PGRX and fastembed-rs Rust implementation of fastembed
I don't know Rust, or Candle, but would love to see more embeddings... embedded.
Flame is named for its ambition to use HuggingFace's Candle library.
see SETUP.md
cargo pgrx run
# [everything just works...]
flame=# CREATE EXTENSION flame;
flame=# SELECT fast_embed('hello world');
- singleton for instatiation
- default model
- tests (long text strings, peformance, multi threading situations, with pgvector)
- choose models at runtime
- auto create pgvector
- more tests