Mathematical Reasoning on AMC 23 (Avg@8, Pass@8, Avg Length)
42.19Avg@8Full-vocab OPD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Full-vocab OPDTeacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=Full-vocab2026.06 | 42.19 | 75 | 4,096 | |
| PowerOPD (alpha=5)Teacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=PowerOPD, alpha=52026.06 | 40.31 | 70 | 2,010 | |
| Vanilla OPD + ClipTeacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=Vanilla, Param=Clip2026.06 | 38.75 | 72.5 | 2,097 | |
| PowerOPD (alpha=10)Teacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=PowerOPD, alpha=102026.06 | 38.75 | 62.5 | 1,997 | |
| PowerOPD (alpha=100)Teacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=PowerOPD, alpha=1002026.06 | 38.75 | 77.5 | 1,821 | |
| PowerOPD (alpha=1)Teacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=PowerOPD, alpha=12026.06 | 37.5 | 70 | 2,113 | |
| Vanilla OPD + Z-ScoreTeacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=Vanilla, Param=Z-Score2026.06 | 36.56 | 67.5 | 2,087 | |
| PowerOPD (alpha=0.5)Teacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=PowerOPD, alpha=0.52026.06 | 36.56 | 67.5 | 2,087 | |
| PowerOPD (alpha=500)Teacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=PowerOPD, alpha=5002026.06 | 36.56 | 77.5 | 1,715 | |
| PowerOPD (alpha=0.1)Teacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=PowerOPD, alpha=0.12026.06 | 36.25 | 65 | 2,142 | |
| Vanilla OPDTeacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=Vanilla2026.06 | 35.94 | 62.5 | 1,881 | |
| Vanilla OPD + TanhTeacher model=Qwen3-8B, Student model=Qwen3-1.7B-Base, Class=Vanilla, Param=Tanh2026.06 | 35.31 | 62.5 | 1,996 |