Mathematical Reasoning on AIME 25 (SC Acc, PoLR Diff, Efficiency, Overhead)
0.767SC AccuracySelf-Consistency
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Self-ConsistencyModel=QWQ32B, Sample size N=512026.01 | 0.767 | — | — | — | |
| PoLRModel=QWQ32B, Sample size N=512026.01 | 0.767 | -10 | 56.8 | 12.3 | |
| Self-ConsistencyModel=QWQ32B, Sample size N=312026.01 | 0.75 | — | — | — | |
| PoLRModel=QWQ32B, Sample size N=312026.01 | 0.75 | -4 | 59.5 | 6.9 | |
| Self-ConsistencyModel=QWQ32B, Sample size N=112026.01 | 0.747 | — | — | — | |
| PoLRModel=QWQ32B, Sample size N=112026.01 | 0.747 | -6 | 65.3 | 5.3 | |
| Self-ConsistencyModel=Phi-4-15B, Sample size N=112026.01 | 0.357 | — | — | — | |
| PoLRModel=Phi-4-15B, Sample size N=112026.01 | 0.357 | 2.3 | 60.5 | 2.3 | |
| Self-ConsistencyModel=DSQ7B, Sample size N=312026.01 | 0.353 | — | — | — | |
| PoLRModel=DSQ7B, Sample size N=312026.01 | 0.353 | 0 | 48.4 | 3.9 | |
| Self-ConsistencyModel=DSQ7B, Sample size N=112026.01 | 0.337 | — | — | — | |
| PoLRModel=DSQ7B, Sample size N=112026.01 | 0.337 | 2.7 | 48.9 | 2.2 | |
| Self-ConsistencyModel=DSQ7B, Sample size N=512026.01 | 0.333 | — | — | — | |
| PoLRModel=DSQ7B, Sample size N=512026.01 | 0.333 | 0 | 48.8 | 7.8 | |
| Self-ConsistencyModel=Phi-4-15B, Sample size N=512026.01 | 0.333 | — | — | — | |
| PoLRModel=Phi-4-15B, Sample size N=512026.01 | 0.333 | 0 | 54.7 | 11 | |
| Self-ConsistencyModel=Phi-4-15B, Sample size N=312026.01 | 0.32 | — | — | — | |
| PoLRModel=Phi-4-15B, Sample size N=312026.01 | 0.32 | 4 | 54.8 | 5.9 |