List your company
ResearchFormerly WizWand

Machine Translation on WMT English-French 2014 (test)

45.6BLEUSOTA (Supervised)
24.38429.89235.440.908Jun 12, 2017Mar 28, 2018Jan 12, 2019Oct 29, 2019Aug 14, 2020May 31, 2021Mar 17, 2022
Updated 8mo ago

Evaluation results

MethodLinks
SOTA (Supervised)Training=Supervised2020.0545.6
mRASPPre-training=mRASP, Fine-tuning=true, Beam Size=52022.0344.3
60L-12L ADMINEvaluation Script=multi-bleu.perl2020.0843.8
Data DiversificationBase Model=Scale Transformer [18]2019.1143.7
CeMATPre-training=CeMAT, Fine-tuning=true, Beam Size=52022.0343.7
Wu et al., 2019bEvaluation Script=multi-bleu.perl2020.0843.3
Wu et al., 2019aEvaluation Script=multi-bleu.perl2020.0843.2
Ott et al., 2018Evaluation Script=multi-bleu.perl2020.0843.2
Dynamic Conv2019.1143.2
Scale Transformer2019.1142.7
TransformerAugmentation=Ens-Distill [7]2019.1142.5
TransformerAugmentation=Distill (T=S) [14]2019.1142.1
Vaswani et al., 2017Evaluation Script=multi-bleu.perl2020.0841.8
60L-12L ADMINEvaluation Script=sacreBLEU.py2020.0841.8
Transformer (big)2017.0641.8
Transformer2019.1141.8
Trans+Rel. Pos2019.1141.5
Ott et al., 2018Evaluation Script=sacreBLEU.py2020.0841.4
DirectTraining=From scratch, Beam Size=52022.0341.4
So et al., 2019Evaluation Script=multi-bleu.perl2020.0841.3
ConvS2S EnsembleTraining Cost (FLOPs)=1.2 * 10^212017.0641.29
GNMT + RL EnsembleTraining Cost (FLOPs)=1.1 * 10^212017.0641.16
mBARTPre-training=mBART, Fine-tuning=true, Beam Size=52022.0341
MoETraining Cost (FLOPs)=1.2 * 10^202017.0640.56
Gehring et al., 2017Evaluation Script=multi-bleu.perl2020.0840.5
ConvS2STraining Cost (FLOPs)=1.5 * 10^202017.0640.46
Deep-Att + PosUnk EnsembleTraining Cost (FLOPs)=8.0 * 10^202017.0640.4
SEQUENCEM=6, N=182021.0440.18
CYCLEM=6, N=182021.0440.17
CYCLE (REV)M=6, N=182021.0440.13
GNMT + RLTraining Cost (FLOPs)=1.4 * 10^202017.0639.92
Deep-Att + PosUnkTraining Cost (FLOPs)=1.0 * 10^202017.0639.2
UniversalM=1, N=62021.0438.84
TransformerAugmentation=Distill (T>S) [14]2019.1138.6
VanillaM=6, N=62021.0438.41
Transformer (base model)2017.0638.1
MASSTraining=Unsupervised2020.0537.5
XLMTraining=Unsupervised2020.0533.4
GPT-3Few-shot learning=Few-shot2020.0532.6
GPT-3Few-shot learning=One-shot2020.0528.3
GPT-3Few-shot learning=Zero-shot2020.0525.2