Machine Translation on IWSLT De-En 14
37.53BLEU ScoreCipherDAug
Evaluation results
| Method | Links | |
|---|---|---|
| CipherDAugkeys=22022.04 | 37.53 | |
| BiBERTType=Existing Method2022.04 | 37.5 | |
| R-DROPType=Existing Method2022.04 | 37.25 | |
| DataDiverseType=Data Augmentation2022.04 | 37.01 | |
| UniDropType=Existing Method2022.04 | 36.88 | |
| MixedRepType=Data Augmentation2022.04 | 36.41 | |
| MATType=Existing Method2022.04 | 36.22 | |
| SeqMixType=Data Augmentation2022.04 | 36.2 | |
| BERT FuseType=Existing Method2022.04 | 36.11 | |
| SwitchOutType=Data Augmentation2022.04 | 35.9 | |
| MixSeqType=Data Augmentation2022.04 | 35.7 | |
| Word DropoutType=Data Augmentation2022.04 | 35.6 | |
| Macaron NetType=Existing Method2022.04 | 35.4 | |
| Transformer2022.04 | 34.71 | |
| LLQROptimizer=AdamW, Block=E-KFAC, Div.=NGD, Backbone=Fairseq Transformer, Time=×1.162026.05 | 34.51 | |
| AdamWOptimizer=AdamW, Backbone=Fairseq Transformer, Time=×1.002026.05 | 34.24 | |
| Transformerbeam=52026.02 | 33.83 | |
| Transformerbeam=12026.02 | 32.89 | |
| SCLMKD=true2026.02 | 32.56 | |
| NeoDiff2026.02 | 32.2 | |
| RDM2026.02 | 32.14 | |
| WDR2026.02 | 31.83 | |
| AR-Diffusion2026.02 | 31.8 | |
| DINOISER2026.02 | 31.61 | |
| SCLMKD=false2026.02 | 31.21 | |
| DifformerKD=true2026.02 | 31.19 | |
| SeqDiffuSeq2026.02 | 30.45 | |
| Compositional CodingM x K=64 x 16, #vectors=1024, vector size=0.89 MB, code len=256 bits, code size=1.22 MB, total size=2.11 MB2017.11 | 29.56 | |
| Baseline NMT#vectors=40000, vector size=35 MB, total size=35 MB2017.11 | 29.45 | |
| GENIE2026.02 | 29.45 | |
| Iterative Pruningpruning ratio=90%, #vectors=40000, vector size=5.21 MB, total size=5.21 MB2017.11 | 29.34 | |
| DiffusionLM2026.02 | 29.11 | |
| Compositional CodingM x K=32 x 16, #vectors=512, vector size=0.44 MB, code len=128 bits, code size=0.61 MB, total size=1.05 MB2017.11 | 29.04 | |
| Iterative Pruningpruning ratio=95%, #vectors=40000, vector size=2.63 MB, total size=2.63 MB2017.11 | 28.84 | |
| DifformerKD=false2026.02 | 28.58 |