Neural Machine Translation on WMT16
32.31BLEUNeoDiff
Evaluation Results
| Method | Links | |
|---|---|---|
| NeoDiffModel type=H, Beam size=102025.05 | 32.31 | |
| DifformerModel type=C, Beam size=102025.05 | 30.93 | |
| NeoDiffModel type=H, Beam size=12025.05 | 30.87 | |
| DifformerModel type=C, Beam size=12025.05 | 30.74 | |
| AbsorbingModel type=D, Beam size=52025.05 | 30.41 | |
| LayerNormNorm Position=Post-Norm2022.10 | 27.3 | |
| LayerNormNorm Position=Pre-Norm2022.10 | 27.3 | |
| Regularized BNNorm Position=Post-Norm2022.10 | 26.5 | |
| Regularized BNNorm Position=Pre-Norm2022.10 | 26.2 | |
| SeqDiffuSeqModel type=C, Beam size=102025.05 | 26.17 | |
| MultinomialModel type=D, Beam size=52025.05 | 25.25 | |
| BatchNormNorm Position=Pre-Norm2022.10 | 25.2 | |
| BatchNormNorm Position=Post-Norm2022.10 | 25 | |
| SeqDiffuSeqModel type=C, Beam size=12025.05 | 23.98 |