Taming the Thinker: Conditional Entropy Shaping for Adaptive LLM Reasoning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wei, Shuyu, Sun, Jian, Qiu, Delai, Wang, Yining, Liu, Shengping, Liang, Jiaen, Fu, Ying, Huang, Wei, Sang, Jitao
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917510204358656
author Wei, Shuyu
Sun, Jian
Qiu, Delai
Wang, Yining
Liu, Shengping
Liang, Jiaen
Fu, Ying
Huang, Wei
Sang, Jitao
author_facet Wei, Shuyu
Sun, Jian
Qiu, Delai
Wang, Yining
Liu, Shengping
Liang, Jiaen
Fu, Ying
Huang, Wei
Sang, Jitao
contents Entropy-based deep reasoning has emerged as a promising direction for improving the reasoning capabilities of Large Language Models (LLMs), but existing methods often either increase response length indiscriminately or shorten responses at the cost of accuracy. To better balance this trade-off, we introduce Conditional Entropy Shaping (CES), a framework that dynamically controls token-level response entropy, enabling LLMs to produce concise solutions on simple problems while encouraging deeper exploration on hard ones. Built on DAPO, CES uses token-level entropy as an uncertainty signal and applies a conditional bidirectional policy: it penalizes high-entropy "forking point" tokens on correct reasoning paths to improve conciseness, and rewards them on incorrect paths to encourage exploration and error correction. We implement CES on DeepSeek-R1-Distill-7B and evaluate it on 12 mathematical benchmarks. CES consistently improves average accuracy while reducing response length relative to DAPO, and supplementary experiments show similar trends on a smaller 1.5B backbone and on out-of-domain benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2605_19358
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Taming the Thinker: Conditional Entropy Shaping for Adaptive LLM Reasoning
Wei, Shuyu
Sun, Jian
Qiu, Delai
Wang, Yining
Liu, Shengping
Liang, Jiaen
Fu, Ying
Huang, Wei
Sang, Jitao
Computation and Language
Entropy-based deep reasoning has emerged as a promising direction for improving the reasoning capabilities of Large Language Models (LLMs), but existing methods often either increase response length indiscriminately or shorten responses at the cost of accuracy. To better balance this trade-off, we introduce Conditional Entropy Shaping (CES), a framework that dynamically controls token-level response entropy, enabling LLMs to produce concise solutions on simple problems while encouraging deeper exploration on hard ones. Built on DAPO, CES uses token-level entropy as an uncertainty signal and applies a conditional bidirectional policy: it penalizes high-entropy "forking point" tokens on correct reasoning paths to improve conciseness, and rewards them on incorrect paths to encourage exploration and error correction. We implement CES on DeepSeek-R1-Distill-7B and evaluate it on 12 mathematical benchmarks. CES consistently improves average accuracy while reducing response length relative to DAPO, and supplementary experiments show similar trends on a smaller 1.5B backbone and on out-of-domain benchmarks.
title Taming the Thinker: Conditional Entropy Shaping for Adaptive LLM Reasoning
topic Computation and Language
url https://arxiv.org/abs/2605.19358