Sequence Classification and Regression on GLUE
82.59RTE AccuracyGRAIN
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GRAINBackbone=RoBERTa-large2026.06 | 82.59 | 87.84 | 85.04 | 91.56 | |
| GRAINBackbone=Llama-3.2-1B2026.06 | 82.22 | 85.21 | 84.32 | 90.78 | |
| PCGradBackbone=RoBERTa-large2026.06 | 81.05 | 86.67 | 83.22 | 91.04 | |
| FFTBackbone=Llama-3.2-1B2026.06 | 79.49 | 86.69 | 73.98 | 90.12 | |
| SWABackbone=RoBERTa-large2026.06 | 78.45 | 88.19 | 80.36 | 91.18 | |
| FocalLossBackbone=RoBERTa-large2026.06 | 77.29 | 86.52 | 81.22 | 88.61 | |
| FFTBackbone=RoBERTa-large2026.06 | 76.17 | 84.9 | 83.2 | 91.86 | |
| SWABackbone=Llama-3.2-1B2026.06 | 76.17 | 67.29 | 82.62 | 85.14 | |
| SAMBackbone=Llama-3.2-1B2026.06 | 75.27 | 82.49 | 70.07 | 89.21 | |
| NoisyTuneBackbone=Llama-3.2-1B2026.06 | 74.69 | 84.76 | 67.64 | 83.97 | |
| PCGradBackbone=Llama-3.2-1B2026.06 | 73.88 | 82.84 | 67.77 | 90.11 | |
| NoisyTuneBackbone=RoBERTa-large2026.06 | 70.65 | 86.91 | 71.04 | 91.38 | |
| SAMBackbone=RoBERTa-large2026.06 | 69.27 | 74.83 | 80.61 | 87.58 | |
| FocalLossBackbone=Llama-3.2-1B2026.06 | 49.39 | 68.38 | 62.17 | 83.66 |