TIGER-Lab/Qwen2.5-Math-7B-CFT

TIGER-Lab/Qwen2.5-Math-7B-CFT is a 7.6 billion parameter mathematical reasoning model developed by TIGER-Lab, built upon the Qwen2.5-Math-7B foundation. It utilizes a novel Critique Fine-Tuning (CFT) approach, teaching the model to critique responses rather than imitate answers, leading to enhanced reasoning. This model excels in mathematical problem-solving, achieving high accuracy on benchmarks like MATH and OlympiadBench with exceptional data efficiency.