Speeding Up the Vision Transformer with BatchNorm | Towards Data Science

How integrating Batch Normalization in an encoder-only Transformer architecture can lead to reduced training time and inference time.