Stanford's transformers course, distilled to a single cheatsheet

What it does This repository hosts a VIP cheatsheet that summarizes the core concepts from Stanford’s CME 295: Transformers and Large Language Models. It covers self-attention, architecture variants, optimization techniques such as Flash Attention and quantization, fine-tuning methods like SFT and LoRA, and downstream applications including RAG, agents, and reasoning models. The material is offered as a downloadable PDF alongside a companion textbook with roughly 600 illustrations spread over 250 pages.

The interesting bit The cheatsheet is available in fifteen languages, from Arabic to Thai, and still manages to squeeze modern techniques—test-time scaling from DeepSeek-R1, mixture-of-experts, and LLM-as-a-judge—into the same visual format as foundational transformer mechanics.

Key highlights

• Covers the full stack: transformer foundations, optimization techniques (distillation, quantization, preference tuning), and application patterns (RAG, agents)
• Localized into 15 languages, including Japanese, Persian, Serbian, and Turkish
• Accompanied by a ~600-illustration, 250-page companion textbook
• Tied to the live Stanford course at cme295.stanford.edu
• Downloadable PDF cheatsheet with a preview image showing the visual layout

Verdict Worth bookmarking if you need a quick visual reference for modern LLM mechanics or are prepping for the Stanford course. Skip it if you are hunting for runnable code or model weights.