Find answers from the community

Updated 3 months ago

Is the Llama Index and Pinecone

Is the Llama Index and Pinecone Integration updated for OpenAI text-embedding-3 models? I can only retrieve the correct nodes using ada (super simple test: directly query a chunk/node in the index and see if the retrieved node is the same), but this fails for text-embedding-3. However, this same test does pass when I use the default in-memory Llama Index vector index. I build the Pinecone index following this doc: https://docs.llamaindex.ai/en/stable/examples/vector_stores/PineconeIndexDemo/?h=pinecone, and change the embedding model using Settings.embed_model=OpenAIEmbedding(model="text-embedding-3-small")
L
R
5 comments
it is updated, but it is also a different model, so results could be different

But also, make sure you are rebuilding/recreating the index if you switch embedding models
Thanks for your quick response @Logan M ! I do rebuild the index for each embedding model - the key issue here is that the test should be passed 100% regardless of which embedding model/database is used, which is not the case
Why would the test pass regardless of embedding model? Different embedding models could retrieve different chunks
@Logan M Sorry I just saw this - the key is that each test is conducted independently for each embedding model/index.

Let me try to clarify: Suppose I have embedding models f and g that build vector indexes A and B respectively. Let a be a node inside A and b be a node inside B. Then f(a.text)=a, and g(b.text)=b, meaning model f should always retrieve a from index A, and model g should always retrieve b from B. Now suppose *f * is ada and g is text-embedding-3; my experiments show that f(a.text) always retrieves a from A, but g(b.text) never retrieves b from B. This is why I am wondering if llama index has not updated their integration with pinecone for text-embedding-3.
Hmm. I don't think this is an issue with llama-index per say. Everything is updated
Add a reply
Sign up and join the conversation on Discord