Find answers from the community

Updated 9 months ago

Hi guys. Can someone please explain to

Hi guys. Can someone please explain to me the difference of using a vector store, such as Qdrant, instead of using the default VectorStoreIndex? I have more than 200k chunks and it taking too long to search (about 6 seconds). Does the Vector Store Index implement Approximate Nearest Neighbors?

7 comments

LLogan M

The default vector store just does simple pairwise cosine similarity for every node

Qdrant (and other vector dbs), will be more effecient and scalable, using HNSW and whatnot

LLogan M

How are you testing runtime? index.query() includes the retrieval and response synthesis. 6secs is not out of the norm for this

lluana

I am testing only the function calling for the retriever, despite the response synthesis

lluana

Because I have created a customized retriever class, that I passed to the query engine pipeline

LLogan M

Gotcha that makes sense then

LLogan M

I'm not sure if you are using the default or an integration like qdrant, I would expect qdrant to run much faster (assuming you aren't using :memory: mode)

lluana

Yes, I am using the default VectorStoreIndex. I will try Qdrant. Thanks

Add a reply