LlamaIndex
Log in

Find answers from the community

Updated last year

Metadata

At a glance

The community member is looking for a way to perform a pre-retrieval metadata-aware condition query in the Chromadb Vector Store. They mention that they can do some post-retrieval node processing, but any metadata conditioning at that stage will drop the node count, sometimes to 0. The community member asks if llama-index provides any utility to specify metadata conditioning retrieval.

In the comments, another community member asks for more explanation. The second community member provides an example scenario where they have nodes in Chromadb with a date metadata. If they retrieve 10 nodes based on similarity_top_k and then drop some of them in the post-retrieval phase due to an incorrect date span, they may end up with less than 10 nodes. The community member suggests that if they could query the vector store with a metadata condition, they would always get 10 nodes within the specified date span.

HHoaz
·
Hi everyone. I need a pre retrieval metadata aware condition query in Chromadb Vector Store. Does llama-index provide any utility to specify metadata conditioning retrieval? I can do some post retrieval node processing but in this stage any metadata conditioning will drop the nodes count and sometimes to 0.
L
H
2 comments
LLogan M
Can you explain a bit more what this means?
HHoaz
Lets say all nodes in my chromadb have a date metadata. If I retrieve 10 nodes based of similarity_top_k and then drop some of them in post retrieval phase, because incorrect date span than I will end up with less nodes than 10. But if I could somehow query the vector store with metadata condition so that it would return 10 nodes based on their matadata within specific date span. This way I'll always get 10 nodes in the specified date span.
Add a reply
Sign up and join the conversation on Discord