Natural Language Inference on MultiNLI matched (dev)
88.4AccuracyaMLP_large
Evaluation results
| Method | Links | |
|---|---|---|
| aMLP_largeAttn Size=128, Params (M)=3162021.05 | 88.4 | |
| gMLP_xlargeParams (M)=9412021.05 | 87.7 | |
| BERT_large (ours)Attn Size=1024 (64 x 16), Params (M)=3362021.05 | 87 | |
| BERT_large [2]Attn Size=1024 (64 x 16), Params (M)=3362021.05 | 86.6 | |
| gMLP_largeParams (M)=3652021.05 | 86.2 | |
| aMLP_baseAttn Size=64, Params (M)=1092021.05 | 85.9 | |
| BERT_base (ours)Attn Size=768 (64 x 12), Params (M)=1102021.05 | 85.6 | |
| BERT-base + PoEDebiasing Strategy=Z-Aug, Training Data=Z(D_G* | D_MNLI)2022.03 | 85.22 | |
| BERT-baseDebiasing Strategy=Z-Aug, Training Data=Z(D_G* | D_MNLI)2022.03 | 84.72 | |
| Regularized-confTraining Data=MNLI2022.03 | 84.6 | |
| PoETraining Data=MNLI2022.03 | 84.58 | |
| BERT_base [2]Attn Size=768 (64 x 12), Params (M)=1102021.05 | 84.4 | |
| BERT-base + PoETraining Data=D_MNLI2022.03 | 84.39 | |
| BERT-baseTraining Data=D_MNLI baseline2022.03 | 83.87 | |
| gMLP_baseParams (M)=1302021.05 | 83.7 | |
| BERT-base Main PoE+CETraining Data=MNLI2022.03 | 83.32 | |
| BERT-baseDebiasing Strategy=Seq-Z, Training Data=Z(D_G* | Z(D_MNLI))2022.03 | 82.55 | |
| BERT-baseDebiasing Strategy=Par-Z, Training Data=Z(D_MNLI) U Z(D_G*)2022.03 | 82.48 | |
| Learned-MixinTraining Data=MNLI2022.03 | 80.5 | |
| ESIM GloVeAuxiliary Information=GloVe vectors2017.06 | 74.63 | |
| ESIM dictionaryAuxiliary Information=dictionary definitions2017.06 | 71.39 | |
| ESIM spellingAuxiliary Information=spelling2017.06 | 69.76 | |
| ESIM baselineAuxiliary Information=none2017.06 | 69.05 |