Skip to main content
GGenerative AI
    • Start here
    • Learning paths5 paths
    • Modules11 modules
    • Projects5 projects
    • Quizzes
    • Flashcards
    • Homework
    • Interview prep8 question banks
    • Cheat sheets11 sheets
    • Glossary239 terms
    • Code snippets
    • Ask the mentor
    • My notes
    • My progress
SearchStart here
Menu
    • Start here
    • Learning paths5 paths
    • Modules11 modules
    • Projects5 projects
    • Quizzes
    • Flashcards
    • Homework
    • Interview prep8 question banks
    • Cheat sheets11 sheets
    • Glossary239 terms
    • Code snippets
    • Ask the mentor
    • My notes
    • My progress
  1. Reference
  2. Decode

Glossary

Decode

The phase after prefill where one token is generated per step. Each step has to read the weights and the KV cache, so it is usually limited by memory bandwidth.

Where it is explained

  • How LLMs workLLM foundations
  • LLM inference and servingProduction: cost, serving and observability

See also

Terms explained in the same lessons.

  • Grouped-query attention (GQA)
  • KV cache
  • Prefill
  • Autoregressive generation
  • Causal mask
  • Chunked prefill
Previous termCross-encoderNext termDefence in depth

All terms, A to Z

GGenerative AI

Learn generative AI from the first token to production systems.

  • Privacy
  • Search