04 · How Models Run

Quantization

Lower-precision compression

Storing a model's numbers at lower precision (fewer bits each) to shrink memory and speed up inference, usually with minor quality loss. The bridge that lets big models run on small hardware.

Concrete example

Shrinking a model so it fits on a laptop or phone, trading a sliver of accuracy for a big drop in size.

Why it matters

How capable models get small enough to run locally and cheaply.