Mathematical Reasoning on MATH 500 (Pass@8, Avg@8, Length)
63.25Avg@8 ScoreFull-vocabulary KL OPD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Full-vocabulary KL OPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base2026.06 | 63.25 | 84 | 4,091 | |
| PowerOPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, alpha=5002026.06 | 61.06 | 75.9 | 999 | |
| PowerOPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, alpha=102026.06 | 59.86 | 76.76 | 1,049 | |
| PowerOPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, alpha=52026.06 | 59.82 | 77.12 | 1,062 | |
| Vanilla OPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, Stabilization=Clip2026.06 | 59.56 | 76.68 | 1,062 | |
| PowerOPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, alpha=1002026.06 | 59.36 | 75.74 | 998 | |
| PowerOPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, alpha=0.52026.06 | 59.29 | 76.9 | 1,052 | |
| PowerOPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, alpha=0.12026.06 | 59.27 | 76.78 | 1,087 | |
| PowerOPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, alpha=12026.06 | 59.04 | 76.9 | 1,083 | |
| Vanilla OPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, Stabilization=Tanh2026.06 | 58.82 | 76.04 | 1,078 | |
| Vanilla OPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base, Stabilization=Z-Score2026.06 | 58.48 | 75.98 | 1,084 | |
| Vanilla OPDTeacher=Qwen3-8B, Student=Qwen3-1.7B-Base2026.06 | 57.55 | 75.88 | 1,050 |