Natural Language Understanding on GLUE (QNLI, MNLI, SST-2, QQP)
62QNLI AccuracyCoA-LoRA
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CoA-LoRASolution=∞, Backbone=RoBERTa-Large, Evaluation Protocol=Fine-tuning2025.09 | 62 | 4.34 | 57 | 59 | 8.89 | 91 | 67 | 1.74 | 52 | 60 | 7.87 | 88 | |
| Shared-LoRASolution=1, Backbone=RoBERTa-Large, Evaluation Protocol=Fine-tuning2025.09 | 60 | 2.9 | 21 | 57 | 8.11 | 35 | 61 | -5.06 | 19 | 53 | -1.18 | 32 | |
| LQ-LoRASolution=6, Backbone=RoBERTa-Large, Evaluation Protocol=Fine-tuning2025.09 | 59 | 2.81 | 108 | 57 | 8.13 | 183 | 64 | 1.54 | 91 | 54 | 0.47 | 172 | |
| QLoRASolution=6, Backbone=RoBERTa-Large, Evaluation Protocol=Fine-tuning2025.09 | 58 | — | 119 | 54 | — | 208 | 63 | — | 97 | 54 | — | 189 |