Chunk

You want to use a node parser but you don't want it to parse nodes?

I don't think I understand the use case (and no, it's not possible this way)

Got it. Hmm. if i already have predefined nodes can i still create the hierarchy somehow?

The issue is that i am losing my metadata (page_ids) when i use the default hierarchical node parser

Thanks for replying so quickly.

I already have something like

document = [page1_chunk1,page1_chunk2,page1_chunk3,page2_chunk1, page3_chunk1]

Would like to leverage llamaindexs hierarchical capability to build hierarchical nodes with this context so maybe I am not looking at the right function

Yea I don't think the hierarchical node parser does what you think it does. It parses nodes into hierarchical chunks according to chunk size (i.e. one top level 2048 size chunk is split into 4 512 sized chunks, which the each one is split into 4 128 sized chunks)

Still not 100% what you are trying to do though 🤔 How do you want retrieval to work?

Oh you're right. Thanks for clarifying

We have two types of retrieval needs

1. Directed, simple questions

What was the revenue of AAPL for 23Q3 from their SEC filings?

1. Broad questions

What was Apple's strategy for growing their brand awareness in Asia? Identify differences in strategy for India, China and Vietnam

I was thinking that hierarchical retrieval will work well for question 2) above. I thought it will recursively summarize nodes and create higher level representations of the data

Maybe what we need is agentic search that is able to break down complex questions?

I think the break down is what you need yup

thank you sir

does llamaindex offer any paid consultations or services? we like the product and sometimes need some help - happy to make it worth ur time as well