PromptForge
AI Glossary

Models & architecture

Quantization

Also known as: quantized, quantisation, gguf

Definition

Quantization reduces the numerical precision of model weights — from 16-bit to 8-bit or 4-bit — shrinking memory use and speeding inference at a modest cost in quality.

A 70-billion-parameter model at 16-bit precision needs roughly 140GB of memory, beyond any consumer GPU. Quantized to 4-bit it needs around 40GB, which brings it within reach of high-end workstations.

Quality loss is smaller than intuition suggests. 8-bit quantization is often indistinguishable from full precision. 4-bit is usually acceptable for general use with some degradation on precise reasoning. Below 4-bit, degradation becomes noticeable.

This is the technology that makes local model running practical, and it is why you see model files labelled Q4, Q5, Q8 and similar.

Related terms

Put this into practice

Understanding the term is step one. Our free courses and tools let you actually use it.