Find answers from the community

Updated 2 years ago

Composable index

At a glance

The community member is building a composable graph to modularly add or remove indices, but is having trouble with the "tree_summarize" feature. Another community member suggests that "tree_summarize" can be passed as an argument to the indexes in the graph, and that the composable graph routes queries to specific indexes using the settings from the query configs. However, the original community member is unsure if this is documented and wants to confirm. There is also a discussion around what happens if two List Indices have the information needed to answer a query, with a suggestion that the response mode can be set in the query config to handle this. The community members also discuss passing a separate dict for query configs rather than using kwargs directly, and one community member suggests checking out a guide on composability in the GPT-Index documentation.

Useful resources
I'm building a composable graph as a hamfisted attempt to modularly add or remove indices.

ComposableGraph doesn't let me do things like "tree_summarize" what traversals exist for that? I would be open to using a different means of composing indices.
L
c
10 comments
You can pass tree_summarize as an argument to the indexes in the graph

All the composable graph does is route your queries to specific indexes, and then those indexes are queried using the settings from the query configs πŸ‘
Can you show me the documentation for that. The last time that I tried. It seemed to tell me that this wasn't an accepted keyword. I can try it again.
Additional question:
what if... two List Indices happen to have the information needed to answer a query, will one be ignored?
Yea inside the query config, you can set the response mode under query_kwargs for that index
Possibly! It depends on what the top level index is πŸ€”
oh so I pass it a separate dict for query configs rather than the kwargs directly!
I'll ask this as a separate question in #❓py-issues-and-help so that it can have a thread for it
Add a reply
Sign up and join the conversation on Discord