Mathematical Reasoning on Hendrycks-MATH (held-out 1,000-question)
71.7Accuracy (Base)R1-Distill-1.5B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| R1-Distill-1.5BBase Model=Qwen2.5-Math-1.5B2025.10 | 71.7 | 69.8 | 94.1 | -8.5 | |
| R1-Distill-14BBase Model=Qwen2.5-14B2025.10 | 67.7 | 72.2 | 97.1 | 15.4 | |
| ORZ-7BBase Model=Qwen2.5-7B2025.10 | 66.9 | 77.7 | 89 | 49.1 | |
| R1-Distill-32BBase Model=Qwen2.5-32B2025.10 | 65.7 | 69.7 | 97.5 | 12.5 | |
| ORZ-32BBase Model=Qwen2.5-32B2025.10 | 65.7 | 80.6 | 95.9 | 49.2 | |
| QwQ-32BBase Model=Qwen2.5-32B2025.10 | 65.6 | 70.9 | 98 | 16.3 | |
| ORZ-1.5BBase Model=Qwen2.5-1.5B2025.10 | 39 | 60.3 | 63.1 | 88.3 | |
| R1-Distill-8BBase Model=Llama-3.1-8B2025.10 | 30.2 | 26.6 | 95.8 | -5.5 | |
| ORZ-0.5BBase Model=Qwen2.5-0.5B2025.10 | 28.8 | 39.2 | 38.5 | 107.4 |