Natural Language Understanding on SuperGLUE (val) (BoolQ, CB, COPA, RTE, WIC)
72.47BoolQ AccuracyLoRA-SAM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LoRA-SAMCost=×2, Backbone=T5-base, Training Mode=Fine-tuning2025.08 | 72.47 | 92.32 | 64.2 | 83.47 | 68.55 | 76.2 | |
| Bi-LoRACost=×1, Backbone=T5-base, Training Mode=Fine-tuning2025.08 | 72.25 | 92.86 | 64.6 | 83.61 | 70.69 | 76.8 | |
| LoRACost=×1, Backbone=T5-base, Training Mode=Fine-tuning2025.08 | 72.2 | 92.86 | 63.8 | 83.1 | 68.6 | 76.11 | |
| Full FTCost=×1, Backbone=T5-base, Training Mode=Fine-tuning2025.08 | 72.19 | 92.26 | 64.67 | 84.48 | 68.34 | 76.39 |