Hours of YouTube and 85 handtyped pages of notes on LLM Inference – Start here
Summary
This repository documents a small LLM inference engine built from scratch in C++ and CUDA, aimed at teaching LLM inference from first principles. It includes over 6 hours of YouTube tutorials and 85+ pages of handwritten notes, with a detailed build guide and a clear path from tokenization to token generation. Licensing separates code (MIT) from the notes (rights reserved), and the project lays out the code structure and workflows for running and experimenting with Llama-3.2 models.