Fine-Tuning with Prompting and Retrieval

[ripe sparrow]

Hi, I was wondering how to format my dataset for Fine-Tuning if the task that I'm trying to do involves prompting and retrieving information from source documents. I would also like to know whether fine-tuning is the right way to go about tailoring the response of the LLM to my use case. Thanks!

[brittle thicket]

personally I would have it return a filename

or pointer

as well as possibly a section / position in the file if it is static

maybe as JSON