Add transformer model support, fix Multi-Head Attention quantization and enhance the pipeline

Add transformer model support, fix Multi-Head Attention quantization and enhance the pipeline #48