Abstract: As large language models (LLMs) continue to scale, they exhibit remarkable emergent capabilities in handling various complex reasoning tasks. Solving specific reasoning tasks typically ...