Multilingual Mathematical Reasoning on MGSM 1.0 (test)
69.6Accuracy (ru)MindMerger-Hard
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MindMerger-HardBackbone=MetaMath-Mistral-7B2024.05 | 69.6 | — | 66.8 | 60 | 62.4 | 52.4 | 48.4 | 57.6 | 66.4 | 71.6 | 76.4 | 52.8 | 67.6 | 63.2 | — | |
| MultiReason-LoraBackbone=MetaMath-Mistral-7B2024.05 | 68.4 | — | 66 | 62.4 | 54.4 | 46.8 | 51.2 | 39.6 | 72 | 70 | 76 | 45.9 | 67 | 60.7 | — | |
| LangBridgeBackbone=MetaMath-Mistral-7B2024.05 | 68.4 | — | 68.8 | 65.6 | 58.4 | 50 | 60 | 47.2 | 68.4 | 65.6 | 65.6 | 52.4 | 65.8 | 61.8 | — | |
| MindMerger-SoftBackbone=MetaMath-Mistral-7B2024.05 | 68.4 | — | 69.6 | 68.8 | 57.2 | 57.6 | 59.6 | 53.2 | 69.2 | 71.6 | 79.2 | 56.8 | 69.1 | 65.4 | — | |
| MindMerger-SoftBackbone=MetaMath-Llama-13B2024.05 | 68 | — | 63.6 | 60.4 | 60 | 55.2 | 59.6 | 56.4 | 65.2 | 69.6 | 68.8 | 57.1 | 65.1 | 62.7 | — | |
| MonoReasonBackbone=MetaMath-Mistral-7B2024.05 | 67.2 | — | 70.8 | 57.2 | 50.8 | 38.4 | 34.8 | 16.8 | 70.4 | 71.2 | 78 | 30 | 66.5 | 55.6 | — | |
| mCoT-7BParameters=7B2024.06 | 66.8 | — | 63.8 | 64.8 | 65.2 | 65.6 | 67.6 | 67.2 | 61.2 | 68.4 | 71.6 | — | — | — | 62.4 | |
| Translate-EnBackbone=MetaMath-Mistral-7B2024.05 | 64.9 | — | 56.7 | 63.1 | 57.2 | 54.6 | 58.7 | 47.7 | 50.4 | 58.6 | 69.7 | 53.7 | 60.1 | 58.2 | — | |
| MultiReason-LoraBackbone=MetaMath-Llama-13B2024.05 | 64.8 | — | 64.4 | 61.2 | 58 | 44 | 49.2 | 40.8 | 64 | 67.6 | 68.4 | 44.7 | 64.1 | 58.2 | — | |
| QAlignBackbone=MetaMath-Llama-13B2024.05 | 64.4 | — | 62.4 | 59.2 | 52.4 | 38.4 | 49.6 | 46 | 62 | 67.2 | 69.2 | 44.7 | 62.4 | 57.1 | — | |
| GPT-4-EnPrompting=2-shot2024.06 | 64 | — | 72 | 70 | 71.6 | 17.6 | 40.4 | 64.4 | 73.6 | 71.2 | 80 | — | — | — | — | |
| MonoReasonBackbone=MetaMath-Llama-13B2024.05 | 62 | — | 63.6 | 56 | 48 | 12 | 8.8 | 6.4 | 64 | 67.2 | 70.8 | 9.1 | 61.7 | 45.9 | — | |
| MultiReason-SFTBackbone=MetaMath-Llama-13B2024.05 | 60.8 | — | 59.2 | 61.6 | 58 | 44.8 | 51.6 | 50.8 | 64.8 | 67.6 | 66.4 | 49.1 | 62.6 | 58.6 | — | |
| MindMerger-HardBackbone=MetaMath-Llama-13B2024.05 | 60.8 | — | 61.6 | 52.8 | 51.6 | 48 | 38.4 | 53.6 | 66.8 | 68.4 | 67.6 | 46.7 | 61.4 | 57 | — | |
| QAlignBackbone=MetaMath-Mistral-7B2024.05 | 60.2 | — | 59.8 | 57.2 | 49.4 | 45.6 | 51.2 | 55.2 | 59.2 | 63.6 | 65.8 | 50.7 | 59.3 | 56.7 | — | |
| LangBridgeBackbone=MetaMath-Llama-13B2024.05 | 58.8 | — | 54.8 | 42 | 33.6 | 39.2 | 42.8 | 42 | 55.2 | 60.8 | 65.2 | 41.3 | 52.9 | 49.4 | — | |
| xCoTParameters=13B2024.06 | 56.8 | — | 46.4 | 54 | 49.6 | 50 | 50 | 51.6 | 52.4 | 54.8 | 54.4 | — | — | — | 47.2 | |
| GPT-3.5-EnPrompting=2-shot2024.06 | 50.4 | — | 59.2 | 52.8 | 46.8 | 7.6 | 15.6 | 40 | 62 | 61.2 | 67.2 | — | — | — | — | |
| xCoTParameters=7B2024.06 | 50 | — | 49.6 | 50 | 50 | 40.4 | 49.2 | 48.4 | 47.2 | 48.8 | 48.4 | — | — | — | 42.8 | |
| MathOctopus-MistralParameters=7B, Backbone=Mistral2024.06 | 49.6 | — | 47.2 | 51.6 | 48 | 44 | 48.8 | 51.6 | 50 | 53.2 | 58.4 | — | — | — | — | |
| PaLM-540BPrompting=few-shot, Parameters=540B2024.06 | 48.4 | — | 46.4 | 46.8 | 40 | 46 | 52.8 | 35.2 | 49.2 | 56.8 | 62.4 | — | — | — | 45.6 | |
| MathOctopusParameters=7B2024.06 | 48.4 | — | 38 | 45.2 | 35.6 | 33.2 | 36.4 | 38.4 | 43.6 | 45.2 | 54.8 | — | — | — | — | |
| MathOctopusParameters=13B2024.06 | 47.6 | — | 49.6 | 51.2 | 39.6 | 42 | 46 | 46 | 49.2 | 53.2 | 51.6 | — | — | — | — | |
| Translate-EnBackbone=MetaMath-Llama-13B2024.05 | 45.6 | — | 54 | 58 | 44.4 | 34.8 | 54 | 44.4 | 53.6 | 62.4 | 70.8 | 44.4 | 55.5 | 52.2 | — | |
| WizardMathParameters=13B2024.06 | 34.4 | — | 42 | 28 | 22 | 6.4 | 5.6 | 5.6 | 40.4 | 45.6 | 52.8 | — | — | — | — | |
| MultiReason-SFTBackbone=MetaMath-Mistral-7B2024.05 | 32 | — | 34.8 | 28.8 | 30.8 | 18.4 | 26.4 | 26.8 | 32.4 | 38 | 39.6 | 23.9 | 33.8 | 30.8 | — | |
| WizardMathParameters=7B2024.06 | 30.8 | — | 30.4 | 22.4 | 24 | 2 | 4 | 3.4 | 30.4 | 34.8 | 47.6 | — | — | — | — | |
| GPT-3Prompting=few-shot2024.06 | 28.4 | — | 37.6 | 40 | 26 | 6.4 | 0.8 | 11.2 | 36 | 40.4 | 53.6 | — | — | — | 0.4 | |
| MT0-3.7B + COT FTEvaluation Setting=Zero-shot, Evaluation Mode=CoT Evaluation, Training Strategy=CoT Fine-tuning, Training Scale=Single language adaptation2023.05 | 10.4 | 7.6 | 15.6 | 11.2 | 11 | — | — | — | — | — | — | — | — | — | — | |
| MT5-3.7B + COT FTEvaluation Setting=Zero-shot, Evaluation Mode=CoT Evaluation, Training Strategy=CoT Fine-tuning, Training Scale=Single language adaptation2023.05 | 6.8 | 3.2 | 9.6 | 6 | 7.6 | — | — | — | — | — | — | — | — | — | — | |
| FLAN-T5-11BEvaluation Setting=Zero-shot, Evaluation Mode=CoT Evaluation2023.05 | 5.2 | 0 | 13.2 | 0 | 0 | — | — | — | — | — | — | — | — | — | — | |
| MT0-3.7BEvaluation Setting=Zero-shot, Evaluation Mode=CoT Evaluation2023.05 | 4.8 | 0 | 7.2 | 1.6 | 2.4 | — | — | — | — | — | — | — | — | — | — | |
| GPT-3 (175B)Evaluation Setting=6-Shot prompt, Evaluation Mode=CoT Evaluation2023.05 | 4.4 | 0 | 10.8 | 6.8 | 0.8 | — | — | — | — | — | — | — | — | — | — | |
| FLAN-T5-3BEvaluation Setting=Zero-shot, Evaluation Mode=CoT Evaluation2023.05 | 2.8 | 0 | 7.2 | 0 | 0 | — | — | — | — | — | — | — | — | — | — | |
| MT5-3.7BEvaluation Setting=Zero-shot, Evaluation Mode=CoT Evaluation2023.05 | 1.2 | 0 | 2 | 0.8 | 0.8 | — | — | — | — | — | — | — | — | — | — |