PATS: Process-Level Adaptive Thinking Mode Switching

TL;DR

This paper introduces PATS, a novel framework enabling large-language models to dynamically switch reasoning strategies at the process level based on task difficulty, improving efficiency and accuracy.

Contribution

It presents a new process-level adaptive reasoning paradigm with integrated reward models and beam search, advancing beyond coarse-grained strategy adjustments.

Findings

Achieves high accuracy on mathematical benchmarks.

Balances performance and computational efficiency.

Demonstrates effectiveness of process-level strategy switching.

Abstract

Current large-language models (LLMs) typically adopt a fixed reasoning strategy, either simple or complex, for all questions, regardless of their difficulty. This neglect of variation in task and reasoning process complexity leads to an imbalance between performance and efficiency. Existing methods attempt to implement training-free fast-slow thinking system switching to handle problems of varying difficulty, but are limited by coarse-grained solution-level strategy adjustments. To address this issue, we propose a novel reasoning paradigm: Process-Level Adaptive Thinking Mode Switching (PATS), which enables LLMs to dynamically adjust their reasoning strategy based on the difficulty of each step, optimizing the balance between accuracy and computational efficiency. Our approach integrates Process Reward Models (PRMs) with Beam Search, incorporating progressive mode switching and bad-step penalty mechanisms. Experiments on diverse mathematical benchmarks demonstrate that our methodology achieves high accuracy while maintaining moderate token usage. This study emphasizes the significance of process-level, difficulty-aware reasoning strategy adaptation, offering valuable insights into efficient inference for LLMs.