Adaptively Sparse Transformers

Benchmark Model Rank Results
machine-translation-on-wmt2014-english-germanAdaptively Sparse Transformer (alpha-entmax)#46BLEU score: 26.93
machine-translation-on-wmt2014-english-germanAdaptively Sparse Transformer (1.5-entmax)#58BLEU score: 25.89
machine-translation-on-wmt2016-romanian-englishAdaptively Sparse Transformer (1.5-entmax)#9BLEU score: 33.1
machine-translation-on-wmt2016-romanian-englishAdaptively Sparse Transformer (alpha-entmax)#11BLEU score: 32.89