Natural Language Inference on MNLI mismatched (val)
75.01AccuracyBERT teacher
Evaluation results
| Method | Links | |
|---|---|---|
| BERT teacherRole=Teacher, Architecture=BERT2023.05 | 75.01 | |
| DTARole=Student, Architecture=TinyBERT2023.05 | 59.94 | |
| DTA with DMURole=Student, Architecture=TinyBERT2023.05 | 59.01 | |
| KD + SmoothingRole=Student, Architecture=TinyBERT2023.05 | 57.71 | |
| KD (standard distillation)Role=Student, Architecture=TinyBERT2023.05 | 57.69 | |
| DMURole=Student, Architecture=TinyBERT2023.05 | 55.86 | |
| TinyBERT baselineRole=Student, Architecture=TinyBERT2023.05 | 55.5 | |
| JTTRole=Student, Architecture=TinyBERT2023.05 | 52.66 | |
| Baseline w/ labelled aug. dataRole=Student, Architecture=TinyBERT2023.05 | 46.47 |