Machine Translation on WMT English-German (EnDe) 2014 1.0 (test)
29.3BLEUOtt et al. model (big)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Ott et al. model (big)Model variant=big2020.01 | 29.3 | — | — | |
| Reversible Transformer (big, 300K steps, no weight sharing)Model variant=big, Training steps=300K, Weight sharing=no2020.01 | 29.1 | 28.9 | 28.4 | |
| Transformer (big)Model variant=big2020.01 | 28.4 | — | — | |
| Reversible Transformer (base, 500K steps, no weight sharing)Model variant=base, Training steps=500K, Weight sharing=no2020.01 | 28 | 27.9 | 27.4 | |
| Reversible Transformer (base, 100K steps)Model variant=base, Training steps=100K2020.01 | 27.6 | 27.4 | 26.9 | |
| Transformer (base model)Model variant=base2020.01 | 27.3 | — | — |