Luning Wang, Larnell Moore, Brandon Zhang, John Kim, Junkuan Liu
This paper introduces a collaborative, training-free, multi-agent reasoning framework that leverages small language models (SLMs) to solve complex mathematical problems efficiently. While large language models (LLMs) such as GPT-4o demonstrate strong reasoning capabilities, their substantial computational demands make them impractical for many real-world applications. We evaluate the effectiveness of this framework on benchmarks including GSM8K and AGIEval-SAT-Math.