Fetching the paper…

Improving Deep Transformer with Depth-Scaled Initialization and Merged Attention · Around