Mathematical Reasoning on AIME24 (accuracy, Average)
16.7AccuracySFT→RL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SFT→RLBackbone=Qwen3-8B-Base2025.09 | 16.7 | 45.5 | |
| BRIDGEBackbone=Qwen3-8B-Base2025.09 | 16.7 | 49.9 | |
| SFTBackbone=Qwen3-8B-Base2025.09 | 13.3 | 37.6 | |
| CHORDBackbone=Qwen3-8B-Base2025.09 | 13.3 | 45.9 | |
| RLBackbone=Qwen3-8B-Base2025.09 | 10 | 42.9 | |
| SFT+RLBackbone=Qwen3-8B-Base2025.09 | 10 | 40.1 | |
| LUFFYBackbone=Qwen3-8B-Base2025.09 | 10 | 44 | |
| SRFTBackbone=Qwen3-8B-Base2025.09 | 6.7 | 39.8 | |
| BaseBackbone=Qwen3-8B-Base2025.09 | 3.3 | 26.6 |