Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning

TL;DR

This paper introduces a multi-stage training approach for large language models that reduces reasoning response length by about 28-40% with minimal accuracy loss, improving efficiency without sacrificing performance.

Contribution

It presents a novel combination of supervised fine-tuning and reinforcement learning with an adaptive length penalty to optimize reasoning efficiency.

Findings

Reduces response length by 28-40% across models

Maintains high accuracy with only 1.6-2.5 point drops

Achieves superior efficiency trade-offs compared to state-of-the-art methods

Abstract

The reasoning capabilities of large language models (LLMs) have improved substantially through increased test-time computation, typically in the form of intermediate tokens known as chain-of-thought (CoT). However, CoT often becomes unnecessarily long, increasing computation cost without actual accuracy gains or sometimes even degrading performance, a phenomenon known as ``overthinking''. We propose a multi-stage efficient reasoning method that combines supervised fine-tuning -- via rejection sampling or reasoning trace reformatting -- with reinforcement learning using an adaptive length penalty. We introduce a lightweight reward function that penalizes tokens generated after the first correct answer but encouraging self-verification only when beneficial. We conduct a holistic evaluation across seven diverse reasoning tasks, analyzing the accuracy-response length trade-off. Our approach reduces response length by an average of 28\% for 8B models and 40\% for 32B models, while incurring only minor performance drops of 1.6 and 2.5 points, respectively. Despite its conceptual simplicity, it achieves a superior trade-off compared to more complex state-of-the-art efficient reasoning methods, scoring 76.6, in terms of the area under the Overthinking-Adjusted Accuracy curve ($\text{AUC}_{\text{OAA}}$) -- 5 points above the base model and 2.5 points above the second-best approach.