Natural Language Understanding on SuperGLUE (SST2, MultiRC, CB subset)
94.81Accuracy (SST2)GREATS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GREATSBackbone=PHI-3, Steps=512, Batch size=32, Selection ratio=25%2026.04 | 94.81 | 88.42 | 93.12 | 92.12 | |
| CoLMBackbone=PHI-3, Steps=512, Batch size=32, Selection ratio=25%2026.04 | 94.72 | 82.99 | 93.05 | 90.25 | |
| UniPROT-PBBackbone=PHI-3, Steps=512, Batch size=32, Selection ratio=25%2026.04 | 94.65 | 88.03 | 94.54 | 92.41 | |
| FTBackbone=PHI-3, Steps=512, Batch size=32, Selection ratio=100%2026.04 | 93.91 | 86.05 | 92.72 | 90.89 | |
| SBERTBackbone=PHI-3, Steps=512, Batch size=32, Selection ratio=25%2026.04 | 90.1 | 82.11 | 87.27 | 86.49 | |
| GradNormBackbone=PHI-3, Steps=512, Batch size=32, Selection ratio=25%2026.04 | 87.94 | 57.54 | 69.1 | 71.53 |