Concept

Quantization

Reducing numerical precision to lower memory use and speed up inference, with possible quality trade-offs.

In practice

A quantized model can run locally with less VRAM.

Why it matters

Use the term as an operating concept, not jargon. Ask what it changes in cost, quality, evidence, latency, security or user experience.

Back to glossary →