Think Twice Before You Write – an Entropy-based Decoding Strategy to Enhance LLM Reasoning

Think Twice Before You Write -- an Entropy-based Decoding Strategy to Enhance LLM Reasoning

Jiashu He, Meizhu Liu, Olaitan P Olaleye, Amit Agarwal, M. Avendi, Yassi Abbasi, Matthew Rowe, Hitesh Laxmichand Patel, Paul Li, Tao Sheng, Sujith Ravi, Dan Roth. Under Review (2026).

Abstract. This paper proposes an entropy-guided decoding framework for improving LLM reasoning. The method identifies high-uncertainty token positions, selectively branches generation around those points, and uses an Entropy After Think stopping criterion to reduce unnecessary rollout computation.

Updated: