What is model quantization in Keras?

Quantization reduces model precision to optimize size and speed.

Weights are converted from float32 to int8 or float16, significantly reducing memory footprint and improving inference speed.