Our new X account is live! Follow @wizwand_team for updates

Machine Translation on WMT English-German 2014 (test)

31.4BLEU

KERMIT Joint → Unidirectional Finetuning

Updated 3d ago

Evaluation Results

Method
KERMIT Joint → Unidirectional Finetuning 2019.06	31.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (Our Implementation) 2019.06	31.2	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Unidirectional (p(y \| x) or p(x \| y)) 2019.06	30.7	-	-	-	-	-	-	-	-	-	-	-	-
Data Diversification 2019.11	30.7	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2019.11	30	-	-	-	-	-	-	-	-	-	-	-	-
Dynamic Conv 2019.11	29.7	-	-	-	-	-	-	-	-	-	-	-	-
CoE-Transformer 2021.07	29.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (big) 2021.07	29.3	-	-	-	-	-	-	-	-	-	-	-	-
Scale Transformer 2019.11	29.3	-	-	-	-	-	-	-	-	-	-	-	-
Trans+Rel. Pos 2019.11	29.2	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2018.09	29.11	-	-	-	-	-	-	-	-	-	-	-	-
Transformer Base & Cutoff (w/ JS loss) 2020.09	29.1	-	-	-	-	-	-	-	-	-	-	-	-
Weighted Transformer 2017.11	28.9	-	-	-	-	-	-	-	-	-	-	-	-
Transformer Base & Cutoff (w/o JS loss) 2020.09	28.9	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2019.11	28.9	-	-	-	-	-	-	-	-	-	-	-	-
RNMT+ 2019.05	28.8	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Joint → Unidirectional Finetuning 2019.06	28.7	-	-	-	-	-	-	-	-	-	-	-	-
RNMT+ 2019.05	28.6	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Joint + Marginal Refining (p(x) and p(y)) 2019.06	28.6	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Joint → Bidirectional Finetuning 2019.06	28.6	-	-	-	-	-	-	-	-	-	-	-	-
DynamicConv 2019.01	28.5	-	-	-	-	-	-	110.9	-	3.9	-	-	-
RNMT+ 2019.05	28.5	-	-	-	-	-	-	-	-	-	-	-	-
Weighted Transformer 2017.11	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2017.11	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2019.05	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2018.09	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Evolved Transformer 2020.09	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Weighted Transformer 2020.09	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Adversarial Training 2020.09	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (big) 2017.06	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2019.11	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2019.11	28.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2018.09	28.36	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (+SRU) 2017.09	28.3	-	-	-	-	-	-	19	2.1	-	-	-	-
Transformer Base (So et al., 2019) 2020.09	28.2	-	-	-	-	-	-	-	-	-	-	-	-
Transformer-DRILL 2019.05	28.1	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Joint → Bidirectional Finetuning 2019.06	28.1	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (base) 2021.07	28.1	-	-	-	-	-	-	-	-	-	-	-	-
Admin 2020.09	27.9	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (+SRU) 2017.09	27.8	-	-	-	-	-	-	22	1.8	-	-	-	-
Transformer (Our Implementation) 2019.06	27.8	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Unidirectional (p(y \| x) or p(x \| y)) 2019.06	27.8	-	-	-	-	-	-	-	-	-	-	-	-
DynamicConv 2019.01	27.7	-	-	-	-	-	-	202.3	-	7.2	-	-	-
Transformer (base) 2017.09	27.6	-	-	-	-	-	-	20	2	-	-	-	-
KERMIT Bidirectional (p(y \| x) and p(x \| y)) 2019.06	27.6	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2019.11	27.6	-	-	-	-	-	-	-	-	-	-	-	-
PostNorm + LayerNorm 2019.10	27.58	-	-	-	-	-	-	-	-	-	-	-	-
PostNorm + FixNorm + ScaleNorm 2019.10	27.57	-	-	-	-	-	-	-	-	-	-	-	-
Transformer-Dual 2019.05	27.5	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (Guo et al.) 2019.01	27.4	-	-	-	-	-	-	-	-	1.6	-	-	-
Blockwise Parallel (Stern et al., 2018) 2019.06	27.4	-	-	-	-	-	-	-	-	-	-	-	-
Insertion Transformer (Stern et al., 2019) 2019.06	27.4	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Joint (p(x, y)) 2019.06	27.4	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2017.11	27.3	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (Li et al.) 2019.01	27.3	-	-	-	-	-	-	-	-	1.3	-	-	-
Transformer 2019.05	27.3	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (Vaswani et al., 2017) 2019.06	27.3	-	-	-	-	-	-	-	-	-	-	-	-
PostNorm + LayerNorm 2019.10	27.3	-	-	-	-	-	-	-	-	-	-	-	-
Transformer 2018.09	27.3	-	-	-	-	-	-	-	-	-	-	-	-
Transformer Base 2020.09	27.3	-	-	-	-	-	-	-	-	-	-	-	-
Transformer (base model) 2017.06	27.3	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Bidirectional (p(y \| x) and p(x \| y)) 2019.06	27.2	-	-	-	-	-	-	-	-	-	-	-	-
PreNorm + FixNorm + ScaleNorm 2019.10	27.07	-	-	-	-	-	-	-	-	-	-	-	-
PreNorm + LayerNorm 2019.10	26.82	-	-	-	-	-	-	-	-	-	-	-	-
ConvS2S Ensemble 2017.06	26.36	-	-	-	-	-	-	-	-	-	-	-	-
GNMT + RL 2019.05	26.3	-	-	-	-	-	-	-	-	-	-	-	-
ConvS2S 2019.05	26.3	-	-	-	-	-	-	-	-	-	-	-	-
GNMT + RL Ensemble 2017.06	26.3	-	-	-	-	-	-	-	-	-	-	-	-
SliceNet-Super 2018.06	26.1	-	-	-	-	-	-	-	-	-	-	-	-
DynamicConv 2019.01	26.1	-	-	-	-	-	-	423	-	15.2	-	-	-
MoE 2017.06	26.03	-	-	-	-	-	-	-	-	-	-	-	-
MoE 2017.11	26	-	-	-	-	-	-	-	-	-	-	-	-
MoE 2019.05	26	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Joint + Marginal Refining (p(x) and p(y)) 2019.06	25.8	-	-	-	-	-	-	-	-	-	-	-	-
KERMIT Joint (p(x, y)) 2019.06	25.6	-	-	-	-	-	-	-	-	-	-	-	-
DenseNMT-En-De-15 2018.06	25.52	-	-	-	-	-	-	-	-	-	-	-	-
SliceNet-Full 2018.06	25.5	-	-	-	-	-	-	-	-	-	-	-	-
Iterative Refinement (Lee et al., 2018) 2019.06	25.5	-	-	-	-	-	-	-	-	-	-	-	-
ConvS2S 2017.11	25.2	-	-	-	-	-	-	-	-	-	-	-	-
NART w/ hints 2019.01	25.2	-	-	-	-	-	-	-	-	22.7	-	-	-
ConvS2S 2018.06	25.16	-	-	-	-	-	-	-	-	-	-	-	-
ConvS2S 2018.09	25.16	-	-	-	-	-	-	-	-	-	-	-	-
ConvS2S 2017.06	25.16	-	-	-	-	-	-	-	-	-	-	-	-
ConvS2S 2019.05	25.1	-	-	-	-	-	-	-	-	-	-	-	-
Lookahead (Adam) 2019.07	24.7	-	-	-	-	-	-	-	-	-	-	-	-
GNMT 2018.06	24.61	-	-	-	-	-	-	-	-	-	-	-	-
WPM-32K 2016.09	24.61	-	-	-	-	-	-	-	-	-	-	-	-
GNMT+RL 2017.11	24.6	-	-	-	-	-	-	-	-	-	-	-	-
GNMT + RL 2019.05	24.6	-	-	-	-	-	-	-	-	-	-	-	-
ADAM 2019.07	24.6	-	-	-	-	-	-	-	-	-	-	-	-
GNMT + RL 2017.06	24.6	-	-	-	-	-	-	-	-	-	-	-	-
Adafactor 2019.07	24.51	-	-	-	-	-	-	-	-	-	-	-	-
Lookahead (Adam without warm-up) 2019.07	24.4	-	-	-	-	-	-	-	-	-	-	-	-
WPM-16K 2016.09	24.36	-	-	-	-	-	-	-	-	-	-	-	-
ENAT Embedding Mapping 2019.01	24.3	-	-	-	-	-	-	-	-	20.4	-	-	-
Mixed Word/Character 2016.09	24.17	-	-	-	-	-	-	-	-	-	-	-	-
Autoregressive (Lee et al.) 2019.01	23.8	-	-	-	-	-	-	54	-	-	-	-	-
ByteNet 2018.09	23.75	-	-	-	-	-	-	-	-	-	-	-	-
ByteNet 2017.06	23.75	-	-	-	-	-	-	-	-	-	-	-	-
ByteNet 2017.11	23.7	-	-	-	-	-	-	-	-	-	-	-	-

Showing 100 of 165 rows