the.ai

Compression / Inference

verified

Post-Training Quantization

Take a trained model and store its weights in fewer bits without retraining anything. It is the cheapest possible compression — minutes on one machine — and for large language models it works well enough that four bits is routine. The difficulty is not the average weight but the rare enormous one.

Viz primitive · budget-splitoutlier-range = 4

outlier-range holds 50% of the budget; rest holds the remaining 50%.

Numeric range consumed by outlier channels against the range left for everything else, in equal units. Drag the outliers up to watch them take the scale — every other value in the group loses precision to them.

4

Reviewed by opendroid · 2026-08-18

  • arXiv:2210.17323 — GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
  • arXiv:2211.10438 — SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models