SignalReasoner: Assessing the Upper Bound of 3B Models for Signal Mathematical Reasoning

SignalReasoner investigates reinforcement fine-tuning strategies to adapt Qwen2.5-3B-Base for graduate-level signal mathematical problems using WirelessMATHBench-XL. While supervised chain-of-thought fine-tuning and reinforcement learning from verifiable rewards have improved LLMs' mathematical reasoning, their application to signal processing remains under-explored. The report examines two specific training paradigms: direct reinforcement learning (RL) on WirelessMATHBench-XL with verifiable rewards and supervised fine-tuning (SFT). This research aims to assess the upper bound of 3B models within this specialized domain.