Quantization

Representing model data with fewer or lower-precision values to reduce storage and memory, with task-dependent quality trade-offs.