Mathematical Reasoning on AIME24, MATH500, Olympiad, AMC, Minerva (test)
33.3AIME24 ScoreDARS-7B-ET-Breadth
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DARS-7B-ET-BreadthBackbone=Qwen2.5-Math-7B, Schedule=Equal-Treat, Breadth-Synergy=true2025.08 | 33.3 | 83.8 | 47.8 | 61.3 | 38.4 | 58.1 | 82.1 | |
| DARS-7B-HW-BreadthBackbone=Qwen2.5-Math-7B, Schedule=Hardness-Weighted, Breadth-Synergy=true2025.08 | 33 | 84.5 | 48.4 | 63 | 36.9 | 58.4 | 83.4 | |
| Breadth-NaiveBackbone=Qwen2.5-Math-7B2025.08 | 30.5 | 83.7 | 47.3 | 61.4 | 37.7 | 57.7 | 79.2 | |
| DARS-7B-HWBackbone=Qwen2.5-Math-7B, Schedule=Hardness-Weighted2025.08 | 30.1 | 83.5 | 47.1 | 59.4 | 37.2 | 57.3 | 83.5 | |
| Depth-NaiveBackbone=Qwen2.5-Math-7B2025.08 | 28 | 83.8 | 46.4 | 59 | 37.3 | 57 | 80.3 | |
| DARS-7B-ETBackbone=Qwen2.5-Math-7B, Schedule=Equal-Treat2025.08 | 26.9 | 83.2 | 46.6 | 57.3 | 38.5 | 57 | 81.7 | |
| RLVR baselineBackbone=Qwen2.5-Math-7B2025.08 | 26.8 | 82.2 | 44.3 | 57.2 | 35.7 | 55.3 | 81.4 | |
| DARS-1.5B-HW-BreadthBackbone=Qwen2.5-Math-1.5B, Schedule=Hardness-Weighted, Breadth-Synergy=true2025.08 | 19.3 | 79 | 42.7 | 51.9 | 31.6 | 52.4 | 82.2 | |
| DARS-1.5B-ET-BreadthBackbone=Qwen2.5-Math-1.5B, Schedule=Equal-Treat, Breadth-Synergy=true2025.08 | 18.6 | 79.4 | 42.9 | 50.6 | 31.7 | 52.5 | 80.8 | |
| Breadth-NaiveBackbone=Qwen2.5-Math-1.5B2025.08 | 18.5 | 77.6 | 41.7 | 49.8 | 31.9 | 51.5 | 79.2 | |
| DARS-1.5B-HWBackbone=Qwen2.5-Math-1.5B, Schedule=Hardness-Weighted2025.08 | 17.7 | 76.4 | 40 | 48.4 | 30.8 | 50 | 82.1 | |
| Depth-NaiveBackbone=Qwen2.5-Math-1.5B2025.08 | 16.5 | 76.2 | 39.9 | 47.9 | 30.9 | 50.1 | 79.9 | |
| DARS-1.5B-ETBackbone=Qwen2.5-Math-1.5B, Schedule=Equal-Treat2025.08 | 15.8 | 76 | 40.9 | 47.2 | 30 | 50 | 81.2 | |
| RLVR baselineBackbone=Qwen2.5-Math-1.5B2025.08 | 14.7 | 75.9 | 39.4 | 47.5 | 31.2 | 49.6 | 79.6 | |
| Base ModelBackbone=Qwen2.5-Math-7B2025.08 | 11.6 | 52.3 | 19.7 | 35.2 | 15.3 | 30.1 | 82.1 | |
| Base ModleBackbone=Qwen2.5-Math-1.5B2025.08 | 4 | 35.1 | 16.2 | 20.8 | 9.5 | 21.1 | 77.9 | |
| DARS-Llama-ET-BreadthBackbone=Llama-3.1-8B, Schedule=Equal-Treat, Breadth-Synergy=true2025.08 | 1.46 | 39.4 | 12 | 13.2 | 20.1 | 22 | 67.2 | |
| DARS-Llama-HW-BreadthBackbone=Llama-3.1-8B, Schedule=Hardness-Weighted, Breadth-Synergy=true2025.08 | 1.11 | 39 | 12 | 13.3 | 19.8 | 21.8 | 68.7 | |
| Breadth-NaiveBackbone=Llama-3.1-8B2025.08 | 0.79 | 34.4 | 9.34 | 12.2 | 19 | 19 | 61.1 | |
| GRPO baselineBackbone=Llama-3.1-8B2025.08 | 0.66 | 29.6 | 7.09 | 10.1 | 15.7 | 15.8 | 56.5 | |
| Depth-NaiveBackbone=Llama-3.1-8B2025.08 | 0.43 | 33.6 | 9.4 | 12.3 | 19.7 | 18.9 | 58.6 | |
| Base ModelBackbone=Llama-3.1-8B2025.08 | 0.23 | 6.13 | 1.54 | 2.76 | 2.72 | 3.25 | 52.7 |