Mathematical Reasoning on DM
51.3AccuracyCRPS
Evaluation Results
| Method | Links | |
|---|---|---|
| CRPSBase Model=LLaMA3-8B, #Samples=60K2026.04 | 51.3 | |
| CRPSBase Model=LLaMA3-8B, #Samples=30K2026.04 | 49.2 | |
| SIGMABase Model=LLaMA3-8B, #Samples=60K2026.04 | 49.2 | |
| CRPSBase Model=Mistral-7B-v0.1, #Samples=60K2026.04 | 48.9 | |
| DARTBase Model=LLaMA3-8B, #Samples=590K2026.04 | 48 | |
| SIGMABase Model=LLaMA3-8B, #Samples=30K2026.04 | 47.5 | |
| CRPSBase Model=Mistral-7B-v0.1, #Samples=30K2026.04 | 46.8 | |
| SIGMABase Model=Mistral-7B-v0.1, #Samples=60K2026.04 | 46.1 | |
| DARTBase Model=Mistral-7B-v0.1, #Samples=590K2026.04 | 45.1 | |
| CRPSBase Model=LLaMA3-8B, #Samples=15K2026.04 | 44.5 | |
| SIGMABase Model=Mistral-7B-v0.1, #Samples=30K2026.04 | 43.8 | |
| MathFusionBase Model=LLaMA3-8B, #Samples=60K2026.04 | 43.4 | |
| MathFusion (Sequential)Base Model=LLaMA3-8B, #Samples=30K2026.04 | 42 | |
| RFTBase Model=LLaMA3-8B, #Samples=590K2026.04 | 41.7 | |
| CRPSBase Model=Mistral-7B-v0.1, #Samples=15K2026.04 | 41.2 | |
| MMIQCBase Model=LLaMA3-8B, #Samples=2.3M2026.04 | 41 | |
| DARTBase Model=LLaMA3-8B, #Samples=60K2026.04 | 39.9 | |
| MathFusion# Samples=60K, Backbone=Mistral-7B2025.03 | 39.2 | |
| MathFusionBase Model=Mistral-7B-v0.1, #Samples=60K2026.04 | 39.2 | |
| WizardMathBase Model=Mistral-7B-v0.1, #Samples=418K2026.04 | 38.4 | |
| MathFusion (DeepSeekMath-7B-RL)# Samples=60K, Backbone=DeepSeekMath-7B-RL2025.03 | 36.5 | |
| DART-Math# Samples=60K, Backbone=Mistral-7B2025.03 | 36 | |
| DARTBase Model=Mistral-7B-v0.1, #Samples=60K2026.04 | 36 | |
| RFTBase Model=Mistral-7B-v0.1, #Samples=590K2026.04 | 35.6 | |
| MetaMathBase Model=LLaMA3-8B, #Samples=400K2026.04 | 35 | |
| MathFusion (Sequential)# Samples=30K, Backbone=Mistral-7B2025.03 | 29.3 | |
| MathFusion (Sequential)Base Model=Mistral-7B-v0.1, #Samples=30K2026.04 | 29.3 | |
| MetaMathBase Model=Mistral-7B-v0.1, #Samples=400K2026.04 | 28 | |
| MetaMathBase Model=Mistral-7B-v0.1, #Samples=60K2026.04 | 27.2 | |
| MathFusion (Parallel)# Samples=30K, Backbone=Mistral-7B2025.03 | 26.5 | |
| MathFusion (Conditional)# Samples=30K, Backbone=Mistral-7B2025.03 | 21.4 | |
| Standard# Samples=15K, Backbone=Mistral-7B2025.03 | 17 | |
| GPT Rewritten# Samples=15K, Backbone=Mistral-7B2025.03 | 13.9 | |
| Mosaic-IT + Original GSM8K and MATH# Samples=30K, Backbone=Mistral-7B2025.03 | 9.8 | |
| Mosaic-IT# Samples=15K, Backbone=Mistral-7B2025.03 | 9.2 |