Mathematical Reasoning on GSM8K (Accuracy & Performance Gain)
91.28AccuracySDAR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SDARBase Model=Qwen3-8B2026.03 | 91.28 | 1.67 | |
| TraDOBase Model=Qwen3-8B2026.03 | 91.05 | 1.44 | |
| Qwen3-8BBase Model=Qwen3-8B2026.03 | 89.61 | — | |
| Qwen2.5-7BBase Model=Qwen2.5-7B2026.03 | 87.87 | — | |
| dTRPOBase Model=Qwen2.5-7B2026.03 | 85.97 | -1.9 |