Mathematical Reasoning on MATH 500 (Accuracy and Token Deltas vs Base)
91.6AccuracyDEER
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DEERBackbone=QwQ-32B, Temperature=0.62025.12 | 91.6 | 0.66 | 3,039.63 | -22 | |
| BaselineBackbone=QwQ-32B, Temperature=0.62025.12 | 91 | — | 3,895.07 | — | |
| LYNXBackbone=QwQ-32B, Conf=0.97, Temperature=0.62025.12 | 89.2 | -1.98 | 3,067.14 | -21.3 | |
| LYNXBackbone=QwQ-32B, Conf=0.95, Temperature=0.62025.12 | 87.8 | -3.52 | 2,811.12 | -27.8 | |
| LYNXBackbone=QwQ-32B, Conf=0.9, Temperature=0.62025.12 | 84 | -7.69 | 2,443.15 | -37.3 | |
| BaselineBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Temperature=0.62025.12 | 81.8 | — | 4,598.35 | — | |
| LYNXBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Conf=0.97, Temperature=0.62025.12 | 80.6 | -1.47 | 3,599.51 | -21.7 | |
| LYNXBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Conf=0.95, Temperature=0.62025.12 | 80 | -2.2 | 3,128.06 | -32 | |
| LYNXBackbone=QwQ-32B, Conf=0.8, Temperature=0.62025.12 | 79.2 | -12.97 | 1,925.46 | -50.6 | |
| LYNXBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Conf=0.9, Temperature=0.62025.12 | 77 | -5.87 | 2,600.05 | -43.5 | |
| ToNBackbone=QwQ-32B, Temperature=0.62025.12 | 71.6 | -21.32 | 2,517.44 | -35.4 | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Temperature=0.62025.12 | 68 | -16.87 | 2,701.16 | -41.3 | |
| LYNXBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Conf=0.8, Temperature=0.62025.12 | 68 | -16.87 | 1,866.73 | -59.4 | |
| LYNXBackbone=QwQ-32B, Conf=0.7, Temperature=0.62025.12 | 66 | -27.47 | 1,365.26 | -64.9 | |
| ToNBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Temperature=0.62025.12 | 59.8 | -26.89 | 3,159.85 | -31.3 | |
| LYNXBackbone=DeepSeek-R1-Distill-Qwen-1.5B, Conf=0.7, Temperature=0.62025.12 | 57.2 | -30.07 | 1,453.57 | -68.4 |