3 ms·
Even in production my guess is most teams would be better off just rolling their own embedding model (huggingface) + caching (redis/rocksdb) + FAISS (nearest ne
by twelfthnight 2y ago
Even in production my guess is most teams would be better off just rolling their own embedding model (huggingface) + caching (redis/rocksdb) + FAISS (nearest neighbor) and be good to go. I suppose there is some expertise needed, but working with a vector database vendor has major drawbacks too.
- hackernoteng 2y agoUsing Postgres with pgvector is trivial and cheap. Its also available on AWS RDS.
- jonplackett 2y agoAlso on supabase!
- danielbln 2y agoOr you just shove it into Postgres + pg_vector and just use the DBMS you already use anyway.