Sharing Attention Weights for Fast Transformer

Tong Xiao, Yinqiao Li, Jingbo Zhu +2

Recently, the Transformer machine translation system has shown strong results by stacking attention layers on both the source and target-language sides. But the inference of this m…