What is model quantization in neural networks?

Quantization reduces model precision to improve speed and reduce size.

It converts weights from float32 to int8 or lower precision formats with minimal accuracy loss.