Stanford CME 295 Transformers & Large Language Models is a study guide and reference resource for learning transformer architectures and large language model concepts.
The repository addresses the need for a consolidated reference covering the breadth of transformer and LLM topics taught in Stanford's CME 295 course. It organizes material across several domains: transformer fundamentals including self-attention and architectural variants, LLM techniques spanning prompting, supervised fine-tuning, parameter-efficient methods like LoRA, preference alignment through RLHF and DPO, and reasoning approaches. The guide also covers practical optimizations such as distributed training, KV caching, and speculative decoding, alongside applications in AI agents and model evaluation.
This resource suits students enrolled in or preparing for the Stanford course, practitioners seeking a structured overview of transformer and LLM concepts, and developers building systems with these models who need quick reference material. The repository provides a cheatsheet format alongside a more comprehensive study guide with detailed illustrations and explanations, allowing users to choose between quick lookup and deeper learning depending on their needs.
The project maintains translations across multiple languages, expanding accessibility beyond English-speaking audiences. Development appears focused on creating and refining educational content rather than building software tooling, with the repository serving as a curated knowledge base tied to an active university course.
