List your company

Mathematical Reasoning on GSM8K DeepSeek-distilled (test)

75.36AccuracyFull Synthetic
33.884844.652455.4266.1876May 10, 2026

Evaluation results

MethodLinks
Full SyntheticBackbone Model=Mistral-v0.2-7B, Selection Ratio=100%2026.0575.36
TMAPBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.0571.34
IFDBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.0571.04
LongestBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.0570.89
DEITABackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.0570.89
RandomBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.0567.48
AlpagasusBackbone Model=Mistral-v0.2-7B, Selection Ratio=30%2026.0567.4
Full SyntheticBackbone Model=Llama2-13B, Selection Ratio=100%2026.0559.51
TMAPBackbone Model=Llama2-13B, Selection Ratio=30%2026.0552.77
DEITABackbone Model=Llama2-13B, Selection Ratio=30%2026.0551.78
LongestBackbone Model=Llama2-13B, Selection Ratio=30%2026.0549.66
AlpagasusBackbone Model=Llama2-13B, Selection Ratio=30%2026.0549.28
IFDBackbone Model=Llama2-13B, Selection Ratio=30%2026.0548.67
RandomBackbone Model=Llama2-13B, Selection Ratio=30%2026.0548.52
BaseBackbone Model=Mistral-v0.2-7B2026.0538.59
BaseBackbone Model=Llama2-13B2026.0535.48