Learning Planning-based Reasoning by Trajectories Collection and Process Reward Synthesizing
This addresses the issue of unreliable and slow reasoning in LLMs for complex tasks, though it is incremental as it builds on existing planning-based and process supervision methods.
The paper tackles the problem of high latency and scalability in improving LLM reasoning by proposing a framework that uses Direct Preference Optimization on collected trajectories with synthesized process rewards, achieving results where a 7B model surpasses GPT-3.5-Turbo on logical reasoning benchmarks.
Large Language Models (LLMs) have demonstrated significant potential in handling complex reasoning tasks through step-by-step rationale generation. However, recent studies have raised concerns regarding the hallucination and flaws in their reasoning process. Substantial efforts are being made to improve the reliability and faithfulness of the generated rationales. Some approaches model reasoning as planning, while others focus on annotating for process supervision. Nevertheless, the planning-based search process often results in high latency due to the frequent assessment of intermediate reasoning states and the extensive exploration space. Additionally, supervising the reasoning process with human annotation is costly and challenging to scale for LLM training. To address these issues, in this paper, we propose a framework to learn planning-based reasoning through Direct Preference Optimization (DPO) on collected trajectories, which are ranked according to synthesized process rewards. Our results on challenging logical reasoning benchmarks demonstrate the effectiveness of our learning framework, showing that our 7B model can surpass the strong counterparts like GPT-3.5-Turbo.