Binary Classification on GLUE (test)
90.66QNLI AccuracyFT Teacher
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FT TeacherStudent Architecture=6-layer, 768-dim2026.02 | 90.66 | — | — | 90.82 | — | |
| TinyBERTStudent Architecture=6-layer, 768-dim, Distillation at pretraining stage=true2026.02 | 90.5 | — | — | 91.6 | — | |
| AD-KDStudent Architecture=6-layer, 768-dim2026.02 | 89.84 | — | — | 91.62 | — | |
| PKDStudent Architecture=6-layer, 768-dim2026.02 | 89.65 | — | — | 90.82 | — | |
| CUDStudent Architecture=6-layer, 768-dim2026.02 | 89.23 | — | — | 91.28 | — | |
| LKDStudent Architecture=6-layer, 768-dim2026.02 | 89.09 | — | — | 90.71 | — | |
| SimPLEbackbone=DeBERTa-large, scale=350M, paradigm=Self-trained, labels=none2023.05 | 85.2 | 81 | 85.5 | 92.8 | 86.1 | |
| SETREDbackbone=DeBERTa-large, scale=350M, paradigm=Self-trained, labels=none2023.05 | 83.4 | 80.5 | 83.9 | 92 | 84.9 | |
| SimPLEbackbone=RoBERTa-large, scale=350M, paradigm=Self-trained, labels=none2023.05 | 83.1 | 80.7 | 83.1 | 91.6 | 84.6 | |
| Dropoutbackbone=DeBERTa-large, scale=350M, paradigm=Self-trained, labels=none2023.05 | 81.1 | 80.5 | 84.1 | 91.8 | 84.4 | |
| SETREDbackbone=RoBERTa-large, scale=350M, paradigm=Self-trained, labels=none2023.05 | 80.5 | 80.5 | 80.8 | 88.3 | 82.5 | |
| Baseline-STbackbone=DeBERTa-large, scale=350M, paradigm=Self-trained, labels=none2023.05 | 79 | 80.2 | 83.4 | 92.1 | 83.7 | |
| Dropoutbackbone=RoBERTa-large, scale=350M, paradigm=Self-trained, labels=none2023.05 | 78.5 | 80.5 | 80.9 | 88.8 | 82.2 | |
| DeBERTa-Supparadigm=Zero-shot, scale=350M, adaptation=Supposition classification2023.05 | 77.3 | 79.9 | 84.5 | 90.1 | 82.9 | |
| Baseline-STbackbone=RoBERTa-large, scale=350M, paradigm=Self-trained, labels=none2023.05 | 74.1 | 80.1 | 81.5 | 88.3 | 81 | |
| DeBERTa-Catparadigm=Zero-shot, scale=350M, adaptation=Concatenation-based2023.05 | 71.6 | 70.5 | 74 | 84.6 | 75.2 | |
| RoBERTa-Supparadigm=Zero-shot, scale=350M, adaptation=Supposition classification2023.05 | 71.5 | 78.6 | 81.2 | 87.7 | 79.8 | |
| UPTparadigm=Few-shot, scale=350M, labels=human-generated2023.05 | 70.1 | 72.1 | 68.9 | 92.9 | 76 | |
| LM-BFFparadigm=Few-shot, scale=350M, labels=human-generated2023.05 | 69.2 | 69.8 | 83.9 | 90.3 | 78.3 | |
| PPTparadigm=Few-shot, scale=350M, labels=human-generated2023.05 | 68.8 | 67.2 | 67.9 | 92.3 | 74.1 | |
| EFLparadigm=Few-shot, scale=350M, labels=human-generated2023.05 | 68 | 67.3 | 85.8 | 90.8 | 78 | |
| FLANparadigm=Few-shot, scale=137B, labels=human-generated2023.05 | 63.3 | 75.9 | 84.5 | 94.6 | 79.6 | |
| PETparadigm=Few-shot, scale=350M, labels=human-generated2023.05 | 61.3 | 67.6 | 65.7 | 91.8 | 71.6 | |
| P-tuningparadigm=Few-shot, scale=350M, labels=human-generated2023.05 | 58.8 | 67.6 | 70.8 | 92.6 | 72.5 | |
| LaMDAparadigm=Few-shot, scale=137B, labels=human-generated2023.05 | 55.7 | 58.9 | 70.8 | 92.3 | 69.4 |