Math Reasoning on Minerva (Accuracy)
40.8AccuracySFT
Evaluation Results
| Method | Links | |
|---|---|---|
| SFTBase Model=Qwen2.5-Math-7B2025.07 | 40.8 | |
| Prefix-RFTBase Model=Qwen2.5-Math-7B2025.07 | 40.3 | |
| SFT+RFTBase Model=Qwen2.5-Math-7B2025.07 | 39.7 | |
| RFTBase Model=Qwen2.5-Math-7B2025.07 | 39.3 | |
| LUFFYBase Model=Qwen2.5-Math-7B2025.07 | 37.5 | |
| RL w/ SFT LossBase Model=Qwen2.5-Math-7B2025.07 | 34.9 | |
| Oat-ZeroBase Model=Qwen2.5-Math-7B2025.07 | 34.6 | |
| UFTBase Model=Qwen2.5-Math-7B2025.07 | 33.8 | |
| ReLIFTBase Model=Qwen2.5-Math-7B2025.07 | 33.8 | |
| SimpleRL-ZeroBase Model=Qwen2.5-Math-7B2025.07 | 25 | |
| Qwen2.5-Math-7B2025.07 | 7.4 |