Entropy-Tree: Tree-Based Decoding with Entropy-Guided Exploration

TL;DR

Entropy-Tree is a novel tree-based decoding method for large language models that uses entropy to guide exploration, improving reasoning accuracy and uncertainty calibration over traditional sampling methods.

Contribution

It introduces entropy-guided branching in decoding, unifying exploration efficiency with improved uncertainty estimation in language model reasoning.

Findings

Achieves better pass@k than Multi-chain across models and datasets.

Demonstrates superior AUROC for predictive entropy compared to traditional metrics.

Improves reasoning accuracy and calibration in large language models.

Abstract

Large language models achieve strong reasoning performance, yet existing decoding strategies either explore blindly (random sampling) or redundantly (independent multi-sampling). We propose Entropy-Tree, a tree-based decoding method that exploits entropy as a signal for branching decisions--expanding the search tree only at positions where the model exhibits genuine uncertainty. Entropy-Tree shows superior accuracy and calibration in reasoning tasks: it achieves better pass@k than Multi-chain across multiple models and datasets, and its predictive entropy demonstrates better AUROC compared to several traditional metrics. Entropy-Tree unifies efficient structured exploration and reliable uncertainty estimation within a single decoding procedure.