List your company
ResearchFormerly WizWand

Machine Translation on WMT19 Zh-En (test)

27.25BLEUNMT + MSO
23.776424.678225.5826.4818May 24, 2021Jul 10, 2021Aug 27, 2021Oct 14, 2021Nov 30, 2021Jan 17, 2022Mar 6, 2022
Updated 8mo ago

Evaluation results

MethodLinks
NMT + MSOobjective=Margin-based Sentence-level Objective2021.0527.25
NMT + MTOobjective=Margin-based Token-level Objective2021.0526.94
Transformer + CBMI-adaptiveModel Scale=big2022.0326.3
Transformer + CBMI-adaptiveModel Scale=base2022.0326.21
NMT + LMtraining_mode=joint training2021.0525.9
Transformer + Self-Paced LearningModel Scale=big2022.0325.88
Transformer + BMI-adaptiveModel Scale=big2022.0325.81
Transformer + BMI-adaptiveModel Scale=base2022.0325.77
NMT (Transformer)source=Re-implemented baseline, training steps=300K2021.0525.75
Transformer + Self-Paced LearningModel Scale=base2022.0325.75
Transformer + LM PriorModel Scale=base2022.0325.71
Transformer + Freq-Chi-SquareModel Scale=big2022.0325.64
Transformer + Anti-Focal LossModel Scale=big2022.0325.64
Transformer + Freq-ExponentialModel Scale=big2022.0325.57
Transformer + Focal LossModel Scale=big2022.0325.54
Transformer + Anti-Focal LossModel Scale=base2022.0325.5
TransformerModel Scale=big2022.0325.48
Transformer + Freq-Chi-SquareModel Scale=base2022.0325.43
Transformer + Focal LossModel Scale=base2022.0325.37
TransformerModel Scale=base2022.0325.36
Transformer + Freq-ExponentialModel Scale=base2022.0324.99
Transformer + Simple FusionModel Scale=base2022.0323.91