Mathematical Reasoning on MATH500 (Acc., Len., #Relf./Ans.)
94.8AccuracyPLE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PLEModel=Qwen3-4B, Mode=Think2026.04 | 94.8 | 6,050.45 | 3.56 | |
| HybridModel=Qwen3-4B, Mode=Think2026.04 | 92.02 | 4,679.54 | 19.6 | |
| SFT-onlyModel=Qwen2.5-7B, Mode=Think2026.04 | 86.5 | 4,488.4 | 33.04 | |
| PLEModel=Qwen2.5-7B, Mode=Think2026.04 | 86 | 3,861.34 | 24.64 | |
| PLEModel=Qwen2.5-7B, Mode=No-think2026.04 | 83.2 | 614.46 | 0.01 | |
| HybridModel=Qwen3-4B, Mode=No-think2026.04 | 82.22 | 1,004.13 | 0.12 | |
| PLEModel=Qwen3-4B, Mode=No-think2026.04 | 80.8 | 676.78 | 0.13 | |
| SFT-onlyModel=Qwen2.5-7B, Mode=No-think2026.04 | 63.56 | 593.09 | 0.31 | |
| InstructModel=Qwen2.5-7B, Mode=No-think2026.04 | 59.94 | 703.11 | 0 |