LLM-RL-Visualized

Why we included this project

This repo collects more than a hundred original, hand-drawn diagrams that walk through how large language models and reinforcement learning algorithms work, from transformer internals and attention to SFT, LoRA, DPO, RLHF, PPO, GRPO, and reasoning techniques like CoT and MCTS. It is a study resource, not runnable software, so the value is for anyone trying to understand or teach these systems: the path from MDP fundamentals to modern policy-gradient training is mapped out in one browsable set of SVGs that stay crisp at any zoom. That makes it useful for developers brushing up before a project, students prepping for interviews, or anyone who has to explain RLHF trade-offs to a team without hand-waving. A companion book adds deeper written explanations, and the author keeps correcting and expanding the charts, so the collection tracks newer methods such as GRPO and VLM architectures.