Natural Language Inference on MNLI GLUE matched mismatched average (test)
84.8AccuracyTR-BERT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| TR-BERTBackbone=BERT-base-uncased2022.03 | 84.8 | 1 | — | — | |
| BERTBackbone=BERT-base-uncased2022.03 | 84.2 | 1 | — | — | |
| PoWER-BERTBackbone=BERT-base-uncased2022.03 | 82.9 | 1.1 | — | — | |
| AdapLeRBackbone=BERT-base-uncased2022.03 | 82.9 | 1.42 | — | — | |
| DistilBERTBackbone=BERT-base-uncased2022.03 | 81.8 | 2 | — | — | |
| SDM-ATTACKVictim Model=BERT2023.05 | 78.7 | — | 13.8 | 49 | |
| BERT-AttackVictim Model=BERT2023.05 | 78.1 | — | 14 | 46 | |
| TextFoolerVictim Model=BERT2023.05 | 76.5 | — | 15 | 45 | |
| A2TVictim Model=BERT2023.05 | 66 | — | 14.4 | 45 |