Mathematical Reasoning on MATH 500 (Accuracy and Tokens)
0.868AccuracyCARD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CARDBackbone=Qwen3-1.7B2025.12 | 0.868 | 1,089 | |
| CARDBackbone=DS-R1-Distill-Qwen-1.5B2025.12 | 0.853 | 1,124 | |
| RDoLT (3-tier)Backbone=Qwen3-1.7B2025.12 | 0.852 | 2,156 | |
| ToT (beam-3)Backbone=Qwen3-1.7B2025.12 | 0.849 | 2,047 | |
| Zero-ShotBackbone=Qwen3-1.7B2025.12 | 0.8457 | 1,867 | |
| RDoLT (3-tier)Backbone=DS-R1-Distill-Qwen-1.5B2025.12 | 0.843 | 6,892 | |
| Zero-ShotBackbone=DS-R1-Distill-Qwen-1.5B2025.12 | 0.839 | 2,134 | |
| ToT (beam-3)Backbone=DS-R1-Distill-Qwen-1.5B2025.12 | 0.832 | 8,734 | |
| CoTBackbone=Qwen3-1.7B2025.12 | 0.831 | 1,189 | |
| CoTBackbone=DS-R1-Distill-Qwen-1.5B2025.12 | 0.818 | 6,449 | |
| CARDBackbone=Qwen3-0.6B2025.12 | 0.751 | 1,067 | |
| RDoLT (3-tier)Backbone=Qwen3-0.6B2025.12 | 0.736 | 2,487 | |
| Zero-ShotBackbone=Qwen3-0.6B2025.12 | 0.73 | 1,923 | |
| ToT (beam-3)Backbone=Qwen3-0.6B2025.12 | 0.728 | 3,156 | |
| CoTBackbone=Qwen3-0.6B2025.12 | 0.712 | 1,923 |