Quantization
양자화
A technique that makes an AI model small and light by storing its numbers with less precision. It has nothing to do with quantum computers.
Overview
A model is made of a huge number of numbers (parameters), and writing each one very precisely makes the file big. Quantization writes these numbers a bit less precisely and shorter. That cuts the file size and the memory needed a lot.
Thanks to this, big models can run on an ordinary PC or phone. Models that are shared as files like GGUF, so that anyone can download and use them, are usually quantized. But the more you shrink it, the more the quality of the answers drops, bit by bit.
An everyday example
It is like saving a photo as a smaller file. It gets a little blurrier, but much lighter, so it is easy to send and open.
Why it matters
It is a word you see often as more people run AI directly on their own computers. It helps to know that the same model can differ in speed and smartness depending on how much it was shrunk.
Also calledmodel quantization