Rethinking Perturbations in Encoder-Decoders for Fast Training

Benchmark Model Rank Results
machine-translation-on-iwslt2014-germanTransformer+Rep(Sim)+WDrop#16BLEU score: 36.22Number of Params: 37M
machine-translation-on-wmt2014-english-germanTransformer+Rep(Uni)#3BLEU score: 33.89SacreBLEU: 32.35
text-summarization-on-duc-2004-task-1Transformer+WDrop#1ROUGE-1: 33.06ROUGE-2: 11.45ROUGE-L: 28.51
text-summarization-on-gigawordTransformer+Rep(Uni)#3ROUGE-1: 39.81ROUGE-2: 20.40ROUGE-L: 36.93
text-summarization-on-gigawordTransformer+Wdrop#5ROUGE-1: 39.66ROUGE-2: 20.45ROUGE-L: 36.59