Mathematical Reasoning on AIME 2024 (Reward-weighted Pass@1)
3.45Reward-weighted Pass@1SFT
Evaluation results
| Method | Links | |
|---|---|---|
| SFTModel=Qwen2.5-7B2025.10 | 3.45 | |
| AIRL (Sparse)Model=Qwen2.5-7B2025.10 | 3.43 | |
| AIRL (Step-wise)Model=Qwen2.5-7B2025.10 | 3.41 | |
| AIRL (Interval)Model=Qwen2.5-7B2025.10 | 3.31 | |
| AIRL (Dense)Model=Qwen2.5-7B2025.10 | 3 | |
| SFTModel=Qwen2.5-3B2025.10 | 2.59 | |
| AIRL (Sparse)Model=Qwen2.5-3B2025.10 | 1.21 | |
| AIRL (Dense)Model=Qwen2.5-3B2025.10 | 1.11 | |
| AIRL (Step-wise)Model=Qwen2.5-3B2025.10 | 1.05 | |
| AIRL (Interval)Model=Qwen2.5-3B2025.10 | 0.86 |