Machine Translation on WMT Chinese-English 2019 (test)
27.61BLEUSeq-KD + Ours
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Seq-KD + OursDistillation Type=Sequence-level Knowledge Distillation, Selective Strategy=Enabled2021.05 | 27.61 | 1.88 | |
| Seq-KDDistillation Type=Sequence-level Knowledge Distillation2021.05 | 27.27 | 1.54 | |
| Word-KD + OursDistillation Type=Word-level Knowledge Distillation, Selective Strategy=Enabled2021.05 | 26.62 | 0.89 | |
| Word-KDDistillation Type=Word-level Knowledge Distillation2021.05 | 26.21 | 0.48 | |
| Transformer (Base)Architecture=Transformer (Base)2021.05 | 25.73 | — |