Milvus

DDaBoi

Hi sorry for the noob question. But I have a pipeline that populates a milvus database with embeddings and content already.

I have searched for a few hours but can't seem to find out how I can use llamaindex to use that existing db and existing embeddings as part of a rag workflow.

I am using llama index 0.11.15. I'm creating a milvusvectorstore and whenever I try to create an index from vector store and pass it my milvus object it seems to complain about an openai api key not being present. Which I don't quite understand because I already have all my embeddings pre generated. I may be lost in the sauce here. Any advice? Anyone do this before? Much appreciated, TIA 🙂

10 comments

WWhiteFang_Jr

Hi there,
You can connect to an existing vector store this way:

Plain Text

# Create an index over the documents
from llama_index.core import VectorStoreIndex, StorageContext
from llama_index.vector_stores.milvus import MilvusVectorStore


vector_store = MilvusVectorStore(
    uri="./milvus_demo.db", dim=1536, overwrite=True
)

index = VectorStoreIndex.from_vector_store(
    vector_store=vector_store
)

Just keep in mind that you use the same embedding model that you used to generate embeddings

DDaBoi

So I tried this except it complains about not having an openai key. I attached an image of the error doing basically that exactly except no overwrite

WWhiteFang_Jr

You have your llm and embed_model in place?

DDaBoi

I tried setting the embedding model using Settings.embed_model = llm-embedding-object ( data bricks llm serving endpoint)

WWhiteFang_Jr

OpenAI key are the default ones if llm and embed_model are not found

DDaBoi

It's give mes an assertion error when I try to set the embed_model it says it has been disabled

DDaBoi

Attachment

rn_image_picker_lib_temp_f73ba16a-346e-4a5a-b464-388f4e96e061.jpg

DDaBoi

Oh wait. Maybe I see the error. Is the embedding model a different kind of object to the llm? I just hooked up the llm to two different serving endpoints, one of which happened to be an embedding model

WWhiteFang_Jr

Yes embedding model object is diff from llm object

DDaBoi

Gotcha. Thank you so much!

Add a reply

Find answers from the community

Milvus