Query

At a glance

The community member has a table hosted on Supabase with id, content, metadata, and embeddings, but they did not use LLaMA or Langchain to create it. They are wondering if it is possible to run an index over the table and start querying using LLaMA-index.

The comments indicate that this is not possible as the vector class expects a specific structure. One community member suggests editing the query method in the LLaMA-index Supabase vector store implementation. Another community member has created a search function on Supabase, indexed vectors with HNSW, and can run queries using Langchain's SupabaseVectorStore, but they would like to see similar functionality in LLaMA-index.

The comments also mention that Langchain has the table schema hardcoded, but the community member has created the table to match their schema. Overall, the community members are trying to move their vector stores to work with pre-existing data, but it has been a tedious process.

Useful resources

MMitchMcD

I have a table hosted on supabase with id, content, metadata, embeddings. did not use llama or langchain to create it.
is it possible to run an index over it and start querying using llama-index? thanks!

4 comments

LLogan M

Sadly not, the vector class expects a pretty specific structure.

You'd have to edit the query method here
https://github.com/run-llama/llama_index/blob/2de0badc7e009a8eabe587c8f48a58fae8cd58e9/llama_index/vector_stores/supabase.py#L139

MMitchMcD

argh... i created a search function on supabase, indexed vectors with hnsw, and now can run queries using langchain's

Plain Text

vector_store = SupabaseVectorStore(
    embedding=embeddings,
    client=supabase,
    table_name="documents",
    query_name="match_documents",
)

it'd be great if you could enable smth similar

LLogan M

Langchain also has the table schema hardcoded, but you've created the table to match their schema 😅

LLogan M

but in general yea, we are trying to move our vector stores to be able to work with pre-existing data. Just been a tedious process

Add a reply

Find answers from the community

Query