UD Algorithm

A mixed-precision quantization algorithm that assigns different precisions to different tensors through sensitivity analysis, thereby maintaining the precision of critical paths while compressing the model.