Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement

Benchmark Model Rank Results
math-word-problem-solving-on-mathQwen2.5-Math-1.5B-Instruct(COT,Greedy)–Accuracy: 75.8Parameters (Billions): 1.5
math-word-problem-solving-on-mathQwen2.5-Math-1.5B-Instruct(TIR,Greedy)–Accuracy: 79.9Parameters (Billions): 1.5
math-word-problem-solving-on-mathQwen2.5-Math-72B-Instruct(COT,Greedy)–Accuracy: 85.9Parameters (Billions): 72
math-word-problem-solving-on-mathQwen2.5-Math-72B-Instruct(TIR,Greedy)–Accuracy: 88.1Parameters (Billions): 72
math-word-problem-solving-on-mathQwen2.5-Math-7B-Instruct(COT,Greedy)–Accuracy: 83.6Parameters (Billions): 7
math-word-problem-solving-on-mathQwen2.5-Math-7B-Instruct(TIR,Greedy)–Accuracy: 85.2Parameters (Billions): 7
mathematical-reasoning-on-amc23Qwen2.5-Math-7B-instruct–Acc: 62.5