mrm8488/longformer-base-4096-finetuned-s...

probably youd want to use a vector store index

A summary index will send every chunk to the LLM for every query (which is more geared towards summarizing vs. QA)

I do want to send every chunk to the LLM. The Answer should come directly from the document

as in, an exact textual match (which this model seems to do) with the provided start and end positions

then you just need the chunks associated to your answer with your extra metadata, so @Logan M is 99.9% correct