skip to content

Search
← All concepts

#quantization

Low-bit weights and activations.

2 chapters, oldest era first: one idea traced across the books.

  1. The Engine Room 2019–2023 Ch. 14: PEFT and LoRA Adapters, prompt and prefix tuning, the low-rank update, and QLoRA's NF4.
  2. The Engine Room 2022–2024 Ch. 15: Quantization Int8 outliers, GPTQ's Hessian trick, and AWQ — the math of throwing bits away safely.