TechnologyAugust 28, 2026
quantise
KWAN-taize
Definition
To reduce the precision of numerical values (such as model weights or cache data) by mapping them to a smaller set of discrete, lower-bit integers to save memory and speed up computation.
Etymology
Derived from the noun 'quantum' (from Latin 'quantus', meaning 'how much'), which refers to a discrete, indivisible unit. In technical fields, it evolved to mean the process of approximating continuous values with discrete steps.
In the news
In the article, the author advises developers to quantise the KV cache when deploying models locally, which helps fit large, long-context models into limited GPU memory.
The Six-Day Mystery That Rewrote AI's Price List
Read the full article ↗HackerNoon