Improving Deep Transformer with Depth-Scaled Initialization and Merged Attention.

Biao Zhang, Ivan Titov, Rico Sennrich

Browse the full EMNLP paper archive.