Machine Translation on WMT De-En 14
31.59BLEUDRDA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DRDAgranularity=dynamic2024.06 | 31.59 | — | — | |
| DRDAgranularity=static2024.06 | 31.48 | — | — | |
| Masked Label Smoothing (MLS)Vocabulary Sharing=true2022.03 | 31.43 | — | — | |
| TransformerVocabulary Sharing=false, Label Smoothing=true2022.03 | 31.14 | — | — | |
| TransformerVocabulary Sharing=true, Label Smoothing=false2022.03 | 31.08 | — | — | |
| TransformerVocabulary Sharing=true, Label Smoothing=true2022.03 | 30.98 | — | — | |
| TransformerVocabulary Sharing=false, Label Smoothing=false2022.03 | 30.85 | — | — | |
| Transformer2024.06 | 29.84 | — | — | |
| DINOISERSampling=MBR-52022.12 | 28.8 | — | — | |
| ARBeam width (b)=42018.02 | 28.47 | — | — | |
| FlowSeq-largeKnowledge Distillation=true, Decoding=argmax2019.09 | 28.39 | — | — | |
| ARBeam width (b)=12018.02 | 28.15 | — | — | |
| LD4LG (MT5-base)Sampling=MBR-52022.12 | 27 | — | — | |
| ELF-BSize=105M (+35M)2026.05 | 26.4 | — | — | |
| LD4LG (MT5-base)Sampling=Random2022.12 | 26.2 | — | — | |
| FlowSeq-baseKnowledge Distillation=true, Decoding=argmax2019.09 | 26.16 | — | — | |
| Iterative RefinementRefinement steps (i_dec)=102018.02 | 25.48 | — | — | |
| Iterative RefinementAdaptive refinement steps=true2018.02 | 25.43 | — | — | |
| FlowSeq-largeKnowledge Distillation=false, Decoding=argmax2019.09 | 25.4 | — | — | |
| CDCDSampling=MBR-102022.12 | 25.4 | — | — | |
| ARSize=99M2026.05 | 25.2 | — | — | |
| CDCDSampling=Random2022.12 | 24.9 | — | — | |
| CDCD2026.05 | 24.9 | — | — | |
| E2D2Size=99M2026.05 | 24.8 | — | — | |
| NAT-REGKnowledge Distillation=true, Decoding=argmax2019.09 | 24.77 | — | — | |
| Iterative RefinementRefinement steps (i_dec)=52018.02 | 23.86 | — | — | |
| FlowSeq-baseKnowledge Distillation=false, Decoding=argmax2019.09 | 23.36 | — | — | |
| NATFertility (FT)=true, NPD reranking=100 samples2018.02 | 23.3 | — | — | |
| CMLM-baseKnowledge Distillation=true, Decoding=argmax2019.09 | 22.26 | — | — | |
| NATFertility (FT)=true2018.02 | 21.47 | — | — | |
| NAT w/ FTKnowledge Distillation=true, Decoding=argmax2019.09 | 21.47 | — | — | |
| DuoSize=170M (+35M)2026.05 | 21.3 | — | — | |
| SeqDiffuSeq2026.05 | 21.3 | — | — | |
| Iterative RefinementRefinement steps (i_dec)=22018.02 | 20.39 | — | — | |
| CTC LossKnowledge Distillation=true, Decoding=argmax2019.09 | 19.8 | — | — | |
| CMLM-smallKnowledge Distillation=true, Decoding=argmax2019.09 | 19.26 | — | — | |
| MDLMSize=99M2026.05 | 18.4 | — | — | |
| Diffusion-LMSampling=MBR-52022.12 | 17.3 | — | — | |
| Iterative RefinementRefinement steps (i_dec)=12018.02 | 16.77 | — | — | |
| NAT-IRKnowledge Distillation=true, Decoding=argmax2019.09 | 16.77 | — | — |