EntroCut: Entropy-Guided Adaptive Truncation for Efficient Chain-of-Thought Reasoning in Small-scale Large Reasoning Models

TL;DR

EntroCut is a training-free, entropy-based method that dynamically truncates reasoning in large models, significantly reducing computational costs while maintaining high accuracy.

Contribution

We introduce EntroCut, a novel entropy-guided dynamic truncation technique for efficient reasoning in large models, with a new metric EPR for evaluating efficiency-accuracy trade-offs.

Findings

Reduces token usage by up to 40% with minimal accuracy loss

Outperforms existing training-free truncation methods

Demonstrates practical efficiency improvements in four benchmarks

Abstract

Large Reasoning Models (LRMs) excel at complex reasoning tasks through extended chain-of-thought generation, but their reliance on lengthy intermediate steps incurs substantial computational cost. We find that the entropy of the model's output distribution in early reasoning steps reliably distinguishes correct from incorrect reasoning. Motivated by this observation, we propose EntroCut, a training-free method that dynamically truncates reasoning by identifying high-confidence states where reasoning can be safely terminated. To comprehensively evaluate the trade-off between efficiency and accuracy, we introduce the Efficiency-Performance Ratio (EPR), a unified metric that quantifies relative token savings per unit accuracy loss. Experiments on four benchmarks show that EntroCut reduces token usage by up to 40\% with minimal accuracy sacrifice, achieving superior efficiency-performance trade-offs compared with existing training-free methods. These results demonstrate that entropy-guided dynamic truncation provides a practical approach to mitigate the inefficiency of LRMs.