Multilingual Mathematical Reasoning on MSVAMP
75Accuracy (English)Task Arithmetic
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Task ArithmeticMerged Models=LM + Math2025.02 | 75 | 19.4 | 60.1 | 14.6 | 58 | 65.5 | 60.4 | 41.7 | 59.4 | 38.7 | 49.3 | — | — | |
| w/o MergingBase Model=LM2025.02 | 70.3 | 44.1 | 58.6 | 28.5 | 61.5 | 46.4 | 55.6 | 48.3 | 53.3 | 57.7 | 52.4 | — | — | |
| LED-MergingMerged Models=LM + Math2025.02 | 69 | 41.6 | 56.1 | 39.4 | 62.8 | 59.8 | 59.7 | 51.7 | 61 | 45 | 54.6 | — | — | |
| MonoReasonModel Scale=13B2024.01 | 68.9 | 19.1 | 58.8 | 20.6 | 68.4 | 68.9 | 68.1 | 57 | 67.5 | 20.5 | 51.8 | — | — | |
| MindMerger-Hardstrategy=Hard2024.05 | 67.8 | 50.6 | 60.6 | 43.6 | 65.9 | 65 | 65.6 | 61.4 | 62.1 | 46.7 | 58.9 | 47 | 64.1 | |
| MindMerger-Softstrategy=Soft2024.05 | 67.7 | 54 | 61.7 | 52 | 64.1 | 65 | 64 | 59 | 63.3 | 53.4 | 60.4 | 53.1 | 63.5 | |
| MetaMathModel Scale=13B2024.01 | 67.7 | 17.4 | 56.6 | 14.6 | 67.3 | 65.9 | 64.7 | 57 | 63.7 | 15.7 | 49.1 | — | — | |
| QAlign -> MonoReasonModel Scale=13B2024.01 | 67.6 | 55.2 | 63.8 | 49.2 | 69.5 | 66.4 | 68.1 | 64.3 | 66.4 | 55.5 | 62.6 | — | — | |
| MonoReason2024.05 | 67.1 | 17.2 | 55.9 | 12.5 | 63.9 | 67.1 | 65.2 | 54 | 58.1 | 15.8 | 47.4 | 15.2 | 61.2 | |
| MetaMathModel Scale=7B2024.01 | 65.5 | 16.5 | 53.1 | 14.2 | 61.4 | 61.2 | 60.7 | 53.2 | 58.9 | 17.8 | 46.3 | — | — | |
| MonoReasonModel Scale=7B2024.01 | 65.5 | 15.4 | 54.4 | 15 | 60.9 | 63.3 | 62.2 | 51.9 | 59.3 | 17.1 | 46.2 | — | — | |
| QAlign2024.05 | 65.3 | 54.8 | 55.7 | 41.7 | 62.8 | 63.3 | 63.2 | 58 | 61.1 | 47.7 | 57.2 | 48.1 | 61.3 | |
| QAlign -> MonoReasonModel Scale=7B2024.01 | 65.3 | 54.8 | 55.7 | 41.7 | 62.8 | 63.3 | 63.2 | 58 | 61.1 | 47.7 | 57.2 | — | — | |
| MultiReason-Loratuning=LoRA2024.05 | 60.6 | 39.1 | 55.1 | 39.4 | 60.4 | 56.8 | 59.1 | 55.2 | 64.3 | 43.8 | 53.4 | 40.8 | 58.8 | |
| LangBridge2024.05 | 60.6 | 42.1 | 50.4 | 46.8 | 58.1 | 56.9 | 57 | 45.5 | 55.8 | 46.3 | 52 | 45.1 | 54.9 | |
| Translate-En2024.05 | 60.6 | 43.1 | 55.8 | 47.9 | 43.9 | 51.4 | 50.9 | 50.4 | 53.4 | 51.3 | 50.9 | 47.4 | 52.3 | |
| MathOctopusModel Scale=13B2024.01 | 56.4 | 46.8 | 52 | 35.2 | 47.2 | 53.2 | 48 | 39.2 | 45.6 | 41.2 | 46.5 | — | — | |
| WizardMathModel Scale=13B2024.01 | 56.3 | 12.5 | 37 | 13.7 | 48.7 | 49.4 | 49.4 | 29.5 | 43.8 | 16.3 | 35.7 | — | — | |
| MultiReasonModel Scale=13B2024.01 | 54.5 | 44.6 | 48.1 | 35 | 53.3 | 52.5 | 53.2 | 49.9 | 51.6 | 41.3 | 48.4 | — | — | |
| MAmmoTHModel Scale=13B2024.01 | 53.4 | 12.9 | 47.7 | 5 | 52.3 | 53.9 | 53.8 | 42.2 | 50.7 | 13.7 | 38.6 | — | — | |
| SFTModel Scale=13B2024.01 | 50.9 | 19.8 | 43.3 | 13.9 | 46.2 | 46.1 | 47.8 | 41.8 | 47.8 | 23.4 | 38.1 | — | — | |
| MathOctopusModel Scale=7B2024.01 | 50.7 | 43.4 | 42.6 | 31.8 | 48.4 | 49.4 | 50.6 | 41.1 | 46.9 | 39.3 | 44.1 | — | — | |
| Ties-MergingMerged Models=LM + Math2025.02 | 49.4 | 33.4 | 40 | 29.1 | 45 | 39.6 | 40.1 | 41.2 | 49.5 | 40.7 | 40.8 | — | — | |
| WizardMathModel Scale=7B2024.01 | 48.5 | 10.3 | 36.3 | 16.1 | 39.2 | 44.8 | 37.7 | 37.9 | 37.4 | 17 | 32.5 | — | — | |
| MultiReasonModel Scale=7B2024.01 | 48 | 42.4 | 43.2 | 27.6 | 44.3 | 45.5 | 46.7 | 40.9 | 42.3 | 36.5 | 41.3 | — | — | |
| RFTModel Scale=13B2024.01 | 47.1 | 19.4 | 42.3 | 12.2 | 45.1 | 45.6 | 45.2 | 42.4 | 46.5 | 24.8 | 37.1 | — | — | |
| MultiReason-SFTtuning=SFT2024.05 | 46.8 | 39.8 | 42.9 | 34.8 | 45.6 | 46.1 | 45.8 | 43.4 | 45 | 38.1 | 42.8 | 37.6 | 45.1 | |
| MAmmoTHModel Scale=7B2024.01 | 45.1 | 4.2 | 26.8 | 4.3 | 39.6 | 42.9 | 39.9 | 26.7 | 33.7 | 6.3 | 26.3 | — | — | |
| w/o MergingBase Model=Math2025.02 | 43.2 | 18.6 | 27.1 | 20.4 | 37.8 | 49.5 | 36.6 | 23.8 | 36 | 28.7 | 32.2 | — | — | |
| RFTModel Scale=7B2024.01 | 42.7 | 14.9 | 34.9 | 7.7 | 40.8 | 42.5 | 41.5 | 33.9 | 39.5 | 16.9 | 31.3 | — | — | |
| SFTModel Scale=7B2024.01 | 38.8 | 17.2 | 35.2 | 11.5 | 39 | 39.2 | 39.1 | 31.6 | 39.1 | 18.2 | 30.9 | — | — | |
| BreadcrumbsMerged Models=LM + Math2025.02 | 36.7 | 16.3 | 17.5 | 38.4 | 51.2 | 52.3 | 53.1 | 43.9 | 49.7 | 41 | 40 | — | — | |
| Model StockMerged Models=LM + Math2025.02 | 32.3 | 32.4 | 42.4 | 25.8 | 42.2 | 39.8 | 38.9 | 42.2 | 57.6 | 41 | 39.5 | — | — |