Sharpness-aware Quantization for Deep Neural Networks