Mathematical Reasoning on MATH (Accuracy, Delta)
76.12AccuracyQwen2.5(14B)T
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen2.5(14B)TGeneration Length=80962026.07 | 76.12 | — | |
| RG-OPD (ours)Generation Length=80962026.07 | 52.6 | 11.51 | |
| Reverse KL [1]Generation Length=80962026.07 | 49.8 | 8.71 | |
| Qwen2.5(1.5B)SGeneration Length=80962026.07 | 41.09 | 0 | |
| Qwen2.5(14B)TGeneration Length=10242026.07 | 30.33 | — | |
| RG-OPD (ours)Generation Length=10242026.07 | 28.63 | 1.16 | |
| TSD-KD [10]Generation Length=80962026.07 | 27.89 | -13.2 | |
| Qwen2.5(1.5B)SGeneration Length=10242026.07 | 27.47 | 0 | |
| TSD-KD [10]Generation Length=10242026.07 | 27.1 | -0.37 | |
| TSD-KD [10]*Generation Length=10242026.07 | 26.1 | -1.37 | |
| Reverse KL [1]Generation Length=10242026.07 | 25.08 | -2.39 |