Decrease in the Speed of Quantization Model

@mohit7 My answer at Quantized::cat running time is slower than fp32 model may help answer your question.