Question Answering on MMLU (Accuracy and Token Count)
85.67AccuracyMARS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MARSBackbone=GPT-4o-mini2025.09 | 85.67 | 4,210 | |
| MADBackbone=GPT-4o-mini2025.09 | 85.33 | 8,338 | |
| DMADBackbone=GPT-4o-mini2025.09 | 85.33 | 8,706 | |
| Self-consistencyBackbone=GPT-4o-mini2025.09 | 85 | 1,515 | |
| DMADBackbone=Llama3-70b2025.09 | 84.67 | 9,641 | |
| MARSBackbone=Llama3-70b2025.09 | 84.67 | 4,210 | |
| CoTBackbone=Llama3-70b2025.09 | 84 | 446 | |
| MADBackbone=Llama3-70b2025.09 | 84 | 9,370 | |
| Self-reflectionBackbone=Llama3-70b2025.09 | 83.67 | 1,087 | |
| DMADBackbone=Mixtral 8x22b2025.09 | 83 | 7,570 | |
| Self-consistencyBackbone=Llama3-70b2025.09 | 82.33 | 1,919 | |
| CoTBackbone=GPT-4o-mini2025.09 | 81.67 | 357 | |
| Self-reflectionBackbone=GPT-4o-mini2025.09 | 80.67 | 847 | |
| MARSBackbone=Mixtral 8x22b2025.09 | 77.67 | 3,829 | |
| Self-reflectionBackbone=Mixtral 8x22b2025.09 | 77.33 | 988 | |
| MADBackbone=Mixtral 8x22b2025.09 | 76.67 | 7,200 | |
| Self-consistencyBackbone=Mixtral 8x22b2025.09 | 76.33 | 1,757 | |
| CoTBackbone=Mixtral 8x22b2025.09 | 73.33 | 434 | |
| MADBackbone=GPT-3.52025.09 | 71.33 | 4,647 | |
| DMADBackbone=GPT-3.52025.09 | 71 | 4,023 | |
| MARSBackbone=GPT-3.52025.09 | 71 | 2,692 | |
| Self-reflectionBackbone=GPT-3.52025.09 | 70.33 | 609 | |
| Self-consistencyBackbone=GPT-3.52025.09 | 69.33 | 1,141 | |
| CoTBackbone=GPT-3.52025.09 | 68 | 283 | |
| MARSBackbone=Mixtral 8x7b2025.09 | 63 | 3,887 | |
| CoTBackbone=Mixtral 8x7b2025.09 | 62 | 438 | |
| Self-consistencyBackbone=Mixtral 8x7b2025.09 | 61.67 | 1,742 | |
| MADBackbone=Mixtral 8x7b2025.09 | 61 | 7,118 | |
| Self-reflectionBackbone=Mixtral 8x7b2025.09 | 60 | 1,042 | |
| DMADBackbone=Mixtral 8x7b2025.09 | 54.33 | 8,658 |