Skip to main content

Chapter 62: LLM Architecture & Compute

Learn how LLMs are built and what compute they need. You build an llm-architecture skill to reason about tokenization, context windows, scaling laws, and hardware choices that impact cost and performance.

Goals​

  • Understand tokenization, embeddings, and context windows
  • Compare model families/sizes and their tradeoffs
  • Relate compute (GPU/TPU) choices to training/inference cost
  • Capture architecture/compute considerations in a reusable skill

Lesson Progression​

  • Model anatomy and tokenization
  • Context, attention scaling, and performance tradeoffs
  • Hardware and cost planning for training/inference
  • Finalize the architecture & compute skill

Outcome & Method​

You finish with an architecture/compute reference skill that guides training and deployment choices in later chapters.

Prerequisites​

  • Chapter 61 decision framework