i think the "chain" is confusing. the high-level principle is that we divide up your document into a list of nodes. when you ask a query, we'll feed in the context information from each of the nodes in sequence in order to obtain a final answer
@やまかず do you mean 3 answers? ahhh currently we support outputting a single response per query. adding multiple responses is definitely a TODO.
In the meantime, you could try to manually do this yourself. if you manually query the top-level index and retrieve the relevant nodes, you can then try to retrieve each subindex, and query each subindex for a separate response
Really? Thanks 😆😆😆 I'm glad. Could you please share that code with me...?
I want the top level index to be a summary of each sub-index. (I have the summary in set_text now) I would like to know the "GPT Index" code to manually query the top-level index and get the relevant nodes.
I will then run it on the 3 sub-indexes in a parallel process and get separate responses