模型剪枝量化
- 剪枝:将某些权重变为0量化:用INT8表示FP32的权重/输入(activation)PTQ post training quantization需要确定两个:where s is the scale factor and z is the zero-point - the integer value to which the real value zero is mapped量化权重(已知分布方便量化)量化输入(需要calibration dataset 表示所有输入数据)QAT
复制链接