Mathematical Reasoning on GSM8k (Accuracy, Consistency Rate, Overconfidence Rate)
58.19AccuracyDGAO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DGAOModel=Qwen-2.5-14B2026.05 | 58.19 | 70.02 | 28.27 | |
| SFTModel=Qwen-2.5-14B2026.05 | 57.9 | 69.44 | 28.57 | |
| PAFTModel=Qwen-2.5-14B2026.05 | 49.43 | 69.91 | 36.73 | |
| DGAOModel=Qwen-2.5-3B2026.05 | 42.91 | 68.11 | 28.41 | |
| SFTModel=Qwen-2.5-3B2026.05 | 42.53 | 67.5 | 28.52 | |
| PAFTModel=Qwen-2.5-3B2026.05 | 33.81 | 67.97 | 36.99 | |
| DGAOModel=Llama-3.2-11B2026.05 | 32.49 | 60.35 | 35.02 | |
| SFTModel=Llama-3.2-11B2026.05 | 32.14 | 58.77 | 37.65 | |
| PAFTModel=Llama-3.2-11B2026.05 | 23.96 | 59.84 | 46.51 | |
| DGAOModel=Llama-3.2-3B2026.05 | 23.19 | 58.47 | 34.72 | |
| SFTModel=Llama-3.2-3B2026.05 | 22.85 | 56.78 | 37.56 | |
| DGAOModel=Gemma-3-4B2026.05 | 15.59 | 44.43 | 31.43 | |
| SFTModel=Gemma-3-4B2026.05 | 15.23 | 44.39 | 31.78 | |
| PAFTModel=Llama-3.2-3B2026.05 | 14.36 | 58 | 46.87 | |
| PAFTModel=Gemma-3-4B2026.05 | 12.17 | 48.27 | 38.66 |