Machine Translation on WMT de-en 2016 (test)
32.35SacreBLEUMEGA-softmax
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MEGA-softmaxAttention Type=softmax2022.09 | 32.35 | — | 32.9 | |
| MEGA-laplaceAttention Type=laplace2022.09 | 32.22 | — | 32.81 | |
| XFM-base‡Architecture=Transformer, Note=variant reported in other works2022.09 | 31.33 | — | 31.92 | |
| MEGA-softmaxAttention Type=softmax2022.09 | 28.47 | 29.18 | — | |
| MEGA-laplaceAttention Type=laplace2022.09 | 28.27 | 28.95 | — | |
| XFM-base‡Architecture=Transformer, Note=variant reported in other works2022.09 | 27.33 | 27.97 | — | |
| XFM-baseArchitecture=Transformer2022.09 | — | 27.3 | — |