Mathematical Reasoning on MATH-500 (avg@4, pass@4)
66.6Avg@4 ScoreTeacher (Gemma-3 4B)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Teacher (Gemma-3 4B)Pair=Gemma-2-2B → Gemma-3 4B, Training Protocol=Full Finetune (FFT)2026.05 | 66.6 | 74.8 | |
| Teacher (Qwen3-1.7B)Pair=Qwen2.5-Math-1.5B → Qwen3-1.7B, Training Protocol=Full Finetune (FFT)2026.05 | 65.3 | 77 | |
| OPDPair=Qwen2.5-Math-1.5B → Qwen3-1.7B, Training Protocol=Full Finetune (FFT)2026.05 | 58.2 | 75.4 | |
| ESRPair=Qwen2.5-Math-1.5B → Qwen3-1.7B, Training Protocol=Full Finetune (FFT)2026.05 | 56.2 | 73.8 | |
| Student (Qwen2.5-Math-1.5B)Pair=Qwen2.5-Math-1.5B → Qwen3-1.7B, Training Protocol=Full Finetune (FFT)2026.05 | 50.95 | 72.8 | |
| ESRPair=Gemma-2-2B → Gemma-3 4B, Training Protocol=Full Finetune (FFT)2026.05 | 26.65 | 40.4 | |
| OPDPair=Gemma-2-2B → Gemma-3 4B, Training Protocol=Full Finetune (FFT)2026.05 | 13.9 | 25 | |
| Student (Gemma-2-2B)Pair=Gemma-2-2B → Gemma-3 4B, Training Protocol=Full Finetune (FFT)2026.05 | 13.45 | 28.2 |