List your company
ResearchFormerly WizWand

Machine Translation on WMT17 Zh-En (test)

26.5BLEUStrokeNet
22.96423.88224.825.718Mar 24, 2021Jul 3, 2021Oct 13, 2021Jan 22, 2022May 4, 2022Aug 13, 2022Nov 23, 2022
Updated 8mo ago

Evaluation results

MethodLinks
StrokeNet2022.1126.5
Prior2022.1125.5
Norm2022.1125.3
Vanilla2022.1124.4
Pretrained Transformer LargeInitialization=N/A2021.0324.2
ELUInitialization=Pretrain, Feature Size k (cross)=64, Feature Size k (causal)=64, Train Time (GPU hours)=80h2021.0323.8
T2RInitialization=Pretrain, Feature Size k (cross)=32, Feature Size k (causal)=4, Train Time (GPU hours)=82h2021.0323.8
ELUInitialization=Random Init., Feature Size k (cross)=64, Feature Size k (causal)=64, Train Time (GPU hours)=120h2021.0323.4
RFAInitialization=Random Init., Feature Size k (cross)=32, Feature Size k (causal)=4, Train Time (GPU hours)=135h2021.0323.4
RFAInitialization=Pretrain, Feature Size k (cross)=32, Feature Size k (causal)=4, Train Time (GPU hours)=90h2021.0323.2
T2RInitialization=Random Init., Feature Size k (cross)=32, Feature Size k (causal)=4, Train Time (GPU hours)=123h2021.0323.1