How to build a Cache (CAG) so I can do inference only on curated data?

[rjyandun]

After finishing ch07 about fine-tunning, I wanted to be able to ask questions about a curated set of data.
For example, I have a compilation of data on PDF format that I have gathered over years of work, about 700 files, but is a pain to find answers by hand. How can I create a cache so that I can ask the GPT2 LLM that we just created on ch06/ch07 to find?
But I want the answer to come only from this specific set of files

[rasbt]

I think in this case you are probably interested in using a RAG setup. I have a quick example here: https://github.com/rasbt/RAGs/blob/main/notebooks/mistral-llamaindex-hf-offline.ipynb