Mathematical Reasoning on TheoremQA (Pass@1)
34.1Pass@1TrajFusion
Evaluation Results
| Method | Links | |
|---|---|---|
| TrajFusionBase Model=DeepSeekMath-7B, Train Size=100k2026.02 | 34.1 | |
| DARTMathBase Model=DeepSeekMath-7B, Train Size=590k2026.02 | 32.2 | |
| Vanilla RFTBase Model=DeepSeekMath-7B, Train Size=100k2026.02 | 31.2 | |
| InstructBase Model=DeepSeekMath-7B, Train Size=780k2026.02 | 28.1 | |
| TrajFusionBase Model=DeepSeekMath-7B, Train Size=15k2026.02 | 26.9 | |
| MathFusionBase Model=DeepSeekMath-7B, Train Size=60k2026.02 | 24.6 | |
| ICLBase Model=DeepSeekMath-7B2026.02 | 23.5 | |
| MMIQCBase Model=DeepSeekMath-7B, Train Size=2.3M2026.02 | 23.4 | |
| TrajFusionBase Model=LLaMA3-8B, Train Size=100k2026.02 | 23.3 | |
| Vanilla RFTBase Model=LLaMA3-8B, Train Size=100k2026.02 | 22.6 | |
| Vanilla RFTBase Model=DeepSeekMath-7B, Train Size=15k2026.02 | 21.4 | |
| MathFusionBase Model=LLaMA3-8B, Train Size=60k2026.02 | 20 | |
| TrajFusionBase Model=LLaMA3-8B, Train Size=15k2026.02 | 20 | |
| ICLBase Model=LLaMA3-8B2026.02 | 19.8 | |
| DARTMathBase Model=LLaMA3-8B, Train Size=590k2026.02 | 19.4 | |
| MMIQCBase Model=LLaMA3-8B, Train Size=2.3M2026.02 | 16.2 | |
| Vanilla RFTBase Model=LLaMA3-8B, Train Size=15k2026.02 | 14.9 | |
| MetaMathBase Model=LLaMA3-8B, Train Size=400k2026.02 | 13.8 |