EpochLex entry

Quantization

kwon-tuh-ZAY-shun

Reducing the numerical precision used to represent a model's weights or computations, usually to lower memory use and improve inference efficiency.