I modify this tools made it support the bert model quantize，but after quantize matmul and biasadd，I use int8 pb file replace fp32 pb file，all the output is wrong。