Why do i get different chunks returned when asking the ...

At a glance

The community member posted a question about why they get different chunks returned when asking the same question repeatedly. Other community members responded, suggesting that the issue may be related to the use of a vector store, specifically Chroma DB, which has a "weird retrieval algorithm" and is approximate, leading to slightly different results for similar queries. They recommended exploring other vector databases that may provide more precise results, such as those using HNSW or IVF methods. However, there was no explicitly marked answer provided.

HHarrison

Hi

Quick question

Why do I get different chunks returned when asking the same question over and over again?

6 comments

LLogan M

Are you using a chat engine or agent? What vector store are you using?

HHarrison

possible neither 😄 just the original - index = GPTVectorStoreIndex.from_vector_store(vector_store=vector_store)

query_engine = index.as_query_engine(similarity_top_k=s_topk)

Using chromadb as the vector store

LLogan M

hmm, I know chroma has a very weird retrieval algorithm. Its approximate, so its definitely possible it changes slightly if you've indexed very similar chunks

HHarrison

Thank you - what is a good alternative which is more "precise"

LLogan M

anything using HNSW will be approximate. Other dbs have options for exact match methods like IVF etc.

Tbh though I see the most issues about wonky retrieval from chroma, but never any other vector db

LLogan M

So probably anything else

Add a reply

Find answers from the community

Why do i get different chunks returned when asking the same question over and over again?