Quantization

Representing model weights and activations with fewer bits (e.g., 8-bit integers) to reduce memory and computation.