AI Blog

AI Blog

by Michele Laurelli

Quantization

Technique
Definition

Reducing precision of weights/activations to lower memory and computation.

Converts 32-bit floats to 8-bit integers or lower. Reduces model size 4x with minimal accuracy loss. Essential for edge deployment.

Examples

1

INT8 quantization

2

Mobile deployment

3

4-bit LLM quantization

Michele Laurelli - AI Research & Engineering