Guys! I build and engine that generate text response to chats all from the documents and data that you insert into it! no GPU inference and it is deterministic so it retrieve the evidence with the response. also it can generate hypothesis with the KB you provide! unlike the RAG when you give this more data it works better and not hallucinate! it has 27 nods! and make sure you ingest your pdf or data first it is some how training it! then you can have a chat bot even when there is no internet and no gpu! also you can have a classifier to mix it with a gpu based llm and retrieve real data with this and some other thing with gpu. it is lightweight and only needs cpu! here are the links: https://github.com/insightitsGit/cpullm pip install cpu-llm==1.13.1 https://www.insightits.com/research-tools/cpu-llm feel free to email me if you have any question!