Mathematical Reasoning on GSM8K DeepSeek-distilled (test)
75.36AccuracyFull Synthetic
Evaluation Results
| Method | Links | |
|---|---|---|
| Full SyntheticBackbone Model=Mistral-v0.2-7B, Selection Ratio=100%2026.05 | 75.36 | |
| TMAPBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.05 | 71.34 | |
| IFDBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.05 | 71.04 | |
| LongestBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.05 | 70.89 | |
| DEITABackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.05 | 70.89 | |
| RandomBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.05 | 67.48 | |
| AlpagasusBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.05 | 67.4 | |
| Full SyntheticBackbone Model=Llama2-13B, Selection Ratio=100%2026.05 | 59.51 | |
| TMAPBackbone Model=Llama2-13B, Selection Ratio=30%2026.05 | 52.77 | |
| DEITABackbone Model=Llama2-13B, Selection Ratio=30%2026.05 | 51.78 | |
| LongestBackbone Model=Llama2-13B, Selection Ratio=30%2026.05 | 49.66 | |
| AlpagasusBackbone Model=Llama2-13B, Selection Ratio=30%2026.05 | 49.28 | |
| IFDBackbone Model=Llama2-13B, Selection Ratio=30%2026.05 | 48.67 | |
| RandomBackbone Model=Llama2-13B, Selection Ratio=30%2026.05 | 48.52 | |
| BaseBackbone Model=Mistral-v0.2-7B2026.05 | 38.59 | |
| BaseBackbone Model=Llama2-13B2026.05 | 35.48 |