Mathematical Reasoning on GSM8K (test) (Pass@4, Avg@4)
94.47Pass@4 ScoreGRPO+IMAX
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRPO+IMAXModel=Qwen3-4B2026.05 | 94.47 | 76.14 | |
| GRPO+IMAXModel=Qwen3-8B2026.05 | 94.01 | 64.18 | |
| DAPO+IMAXModel=Qwen3-4B2026.05 | 93.78 | 72.93 | |
| GRPO-prefixModel=Qwen3-4B2026.05 | 93.48 | 71.97 | |
| DAPO-prefixModel=Qwen3-4B2026.05 | 93.18 | 72.02 | |
| MERCI-prefixModel=Qwen3-8B2026.05 | 93.18 | 59.59 | |
| DAPO+IMAXModel=Qwen3-8B2026.05 | 92.72 | 60.2 | |
| GRPO-fullModel=Qwen3-4B2026.05 | 91.81 | 86.64 | |
| GRPO-prefixModel=Qwen3-8B2026.05 | 91.78 | 60.9 | |
| GRPO-fullModel=Qwen3-8B2026.05 | 91.58 | 76.95 | |
| DIVER-prefixModel=Qwen3-4B2026.05 | 91.43 | 66.49 | |
| MERCI-prefixModel=Qwen3-4B2026.05 | 90.83 | 62.51 | |
| DIVER-prefixModel=Qwen3-8B2026.05 | 90.75 | 57.15 | |
| BaseModel=Qwen3-4B2026.05 | 90.6 | 59.7 | |
| Text Init.Model=Qwen3-8B2026.05 | 90.3 | 56.16 | |
| DAPO-prefixModel=Qwen3-8B2026.05 | 90.07 | 52.12 | |
| Text Init.Model=Qwen3-4B2026.05 | 88.55 | 53.49 | |
| BaseModel=Qwen3-8B2026.05 | 88.1 | 55.44 | |
| DAPO-prefixModel=Qwen2.5-1.5B2026.05 | 86.28 | 69.5 | |
| GRPO-prefixModel=Qwen2.5-1.5B2026.05 | 85.75 | 66.89 | |
| GRPO+IMAXModel=Qwen2.5-1.5B2026.05 | 85.75 | 67.4 | |
| DAPO+IMAXModel=Qwen2.5-1.5B2026.05 | 84.99 | 67.63 | |
| GRPO-fullModel=Qwen2.5-1.5B2026.05 | 84.84 | 74.09 | |
| MERCI-prefixModel=Qwen2.5-1.5B2026.05 | 82.11 | 63.34 | |
| Text Init.Model=Qwen2.5-1.5B2026.05 | 80.29 | 56.92 | |
| DIVER-prefixModel=Qwen2.5-1.5B2026.05 | 80.29 | 59.76 | |
| BaseModel=Qwen2.5-1.5B2026.05 | 78.92 | 53.43 |