Question Answering on SQuAD v2 (Accuracy, Consistency, Overconfidence)
87.51AccuracyDGAO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DGAOModel=Qwen-2.5-14B2026.05 | 87.51 | 95.28 | 10.42 | |
| SFTModel=Qwen-2.5-14B2026.05 | 86.92 | 94.9 | 11.16 | |
| DGAOModel=Qwen-2.5-3B2026.05 | 85.63 | 93.35 | 10.22 | |
| SFTModel=Qwen-2.5-3B2026.05 | 84.97 | 92.91 | 11.14 | |
| PAFTModel=Qwen-2.5-14B2026.05 | 83.68 | 95.41 | 13.52 | |
| PAFTModel=Qwen-2.5-3B2026.05 | 81.59 | 93.56 | 13.77 | |
| DGAOModel=Llama-3.2-11B2026.05 | 77.28 | 91.46 | 17.71 | |
| DGAOModel=Gemma-3-4B2026.05 | 75.82 | 85.49 | 14.8 | |
| SFTModel=Llama-3.2-11B2026.05 | 75.65 | 90.95 | 18.61 | |
| DGAOModel=Llama-3.2-3B2026.05 | 75.44 | 89.6 | 17.57 | |
| SFTModel=Gemma-3-4B2026.05 | 74.09 | 85.31 | 16.29 | |
| SFTModel=Llama-3.2-3B2026.05 | 73.73 | 88.98 | 18.54 | |
| PAFTModel=Gemma-3-4B2026.05 | 63.92 | 84.58 | 24.8 | |
| PAFTModel=Llama-3.2-11B2026.05 | 60.84 | 85.12 | 28.76 | |
| PAFTModel=Llama-3.2-3B2026.05 | 58.11 | 82.62 | 29.19 |