Edge-Ai

MXFP4 quantization is a microscaling, 4-bit floating-point compression scheme designed to shrink the memory footprint of deep-learning models without hurting accuracy.