List your company
ResearchFormerly WizWand

Machine Translation on WMT En-Fr 2014

43.8BLEUAdmin
-1.4129610.3250222.06333.80098Apr 17, 2020Aug 11, 2020Dec 5, 2020Apr 1, 2021Jul 26, 2021Nov 19, 2021Mar 16, 2022
Updated 8mo ago

Evaluation results

MethodLinks
AdminParam. #=262 M, dim(W(1)) in FFN=512 × 2048, Enc#-Dec#=60L-12L2020.0443.8
PrimeParam. #=252 M, dim(W(1)) in FFN=1024 × 4096, Enc#-Dec#=6L-6L2020.0443.48
T5-11BParam. #=11 B, dim(W(1)) in FFN=1024 × 65536, Enc#-Dec#=24L-24L2020.0443.4
DG-TransformerParam. #=264 M, dim(W(1)) in FFN=1024 × 4096, Enc#-Dec#=8L-8L2020.0443.27
DynamicConvParam. #=213 M, dim(W(1)) in FFN=1024 × 4096, Enc#-Dec#=7L-7L2020.0443.2
Pre-LNParam. #=262 M, dim(W(1)) in FFN=512 × 2048, Enc#-Dec#=60L-12L2020.0443.1
T5-3BParam. #=3 B, dim(W(1)) in FFN=1024 × 16384, Enc#-Dec#=24L-24L2020.0442.6
T5-LargeParam. #=770 M, dim(W(1)) in FFN=1024 × 4096, Enc#-Dec#=12L-12L2020.0441.5
T5-BaseParam. #=220 M, dim(W(1)) in FFN=512 × 2048, Enc#-Dec#=6L-6L2020.0441.2
Trans.Big-RNMT+Param. #=377 M, dim(W(1)) in FFN=1024 × 8192, Enc#-Dec#=6L-6L2020.0441.12
XLMApproach=+Self-training2022.0339.3
MASSApproach=UNMT2022.0339.2
XLMApproach=UNMT2022.0339.1
MASSApproach=+Self-training2022.0339
CBDImplementation=Existing Works2022.0338.2
MASSImplementation=Our Implementation, Approach=+Self-training2022.0338
XLMImplementation=Our Implementation, Approach=+Self-training2022.0337.8
MASSImplementation=Our Implementation, Approach=UNMT2022.0337.8
MASSImplementation=Existing Works, Approach=UNMT2022.0337.5
XLMImplementation=Our Implementation, Approach=UNMT2022.0337.4
CBDImplementation=Existing Works2022.0335.5
MASSImplementation=Our Implementation, Approach=+Self-training2022.0335.2
XLMImplementation=Our Implementation, Approach=+Self-training2022.0335.1
Supervised Modelmode=supervised2021.0935
MASSImplementation=Existing Works, Approach=UNMT2022.0334.9
MASSImplementation=Our Implementation, Approach=UNMT2022.0334.9
XLMImplementation=Our Implementation, Approach=UNMT2022.0334.5
FLAN 137Bprotocol=zero-shot, template_selection=best dev, number_of_templates=52021.0933.9
FLAN 137Bprotocol=few-shot, template_selection=average, k=9, number_of_templates=52021.0933.9
FLAN 137Bprotocol=few-shot, template_selection=best dev, k=9, number_of_templates=52021.0933.8
XLMImplementation=Existing Works, Approach=UNMT2022.0333.4
XLMImplementation=Existing Works, Approach=UNMT2022.0333.3
FLAN 137Bprotocol=zero-shot, template_selection=average, number_of_templates=52021.0932.9
GPT-3 175Bprotocol=few-shot, k=642021.0932.6
LaMDA-PTprotocol=few-shot, k=52021.0931.5
GPT-3 175Bprotocol=zero-shot2021.0925.2
LaMDA-PTprotocol=zero-shot2021.0911.2
TransEvolve-fullFF-2#Params=59M2021.090.395
Transformer BASE#Params=65M2021.090.381
TransEvolve-fullFF-1#Params=53M2021.090.38
TransEvolve-randomFF-2#Params=33M2021.090.334
TransEvolve-randomFF-1#Params=27M2021.090.326