@inproceedings{cai-etal-2026-entropy,
title = "Entropy Equilibrium Sampling: Tuning-Free Truncation for Language Model Decoding",
author = "Cai, Xiaodong and
Lin, Hai and
Zhan, Shaoxiong and
Luo, Weiqi and
Zheng, Hai-Tao and
Kim, Hong-Gee and
Zhou, XiuTeng and
Ma, Chunxia and
Yang, Yu",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2398/",
pages = "54510--54525",
ISBN = "979-8-89176-459-0",
abstract = "Token sampling strategies critically influence text generation quality in large language models (LLMs). However, existing methods introduce additional hyperparameters, requiring extensive tuning and complicating deployment. We present Entropy Equilibrium Sampling (EES), a tuning-free approach grounded in information theory that dynamically determines the candidate set size using a first-crossing rule: it stops candidate-set expansion at the first index where cumulative probability mass exceeds the normalized entropy of the top-k distribution and retains the preceding prefix. We evaluate EES on reasoning, mathematical, and open-ended generation tasks across multiple model architectures and scales. Our results show that EES achieves robust performance across temperature settings without retuning truncation hyperparameters, delivering competitive accuracy with tuned baselines while maintaining diversity. Analysis reveals that the retained candidate-set size k* automatically encodes temperature effects, adapting from near-greedy behavior at low temperatures to broad candidate retention at high temperatures. By eliminating truncation-specific hyperparameter tuning, EES simplifies deployment while providing stable and competitive performance. Our code is publicly available."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="cai-etal-2026-entropy">
<titleInfo>
<title>Entropy Equilibrium Sampling: Tuning-Free Truncation for Language Model Decoding</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xiaodong</namePart>
<namePart type="family">Cai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hai</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shaoxiong</namePart>
<namePart type="family">Zhan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weiqi</namePart>
<namePart type="family">Luo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hai-Tao</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hong-Gee</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">XiuTeng</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chunxia</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Token sampling strategies critically influence text generation quality in large language models (LLMs). However, existing methods introduce additional hyperparameters, requiring extensive tuning and complicating deployment. We present Entropy Equilibrium Sampling (EES), a tuning-free approach grounded in information theory that dynamically determines the candidate set size using a first-crossing rule: it stops candidate-set expansion at the first index where cumulative probability mass exceeds the normalized entropy of the top-k distribution and retains the preceding prefix. We evaluate EES on reasoning, mathematical, and open-ended generation tasks across multiple model architectures and scales. Our results show that EES achieves robust performance across temperature settings without retuning truncation hyperparameters, delivering competitive accuracy with tuned baselines while maintaining diversity. Analysis reveals that the retained candidate-set size k* automatically encodes temperature effects, adapting from near-greedy behavior at low temperatures to broad candidate retention at high temperatures. By eliminating truncation-specific hyperparameter tuning, EES simplifies deployment while providing stable and competitive performance. Our code is publicly available.</abstract>
<identifier type="citekey">cai-etal-2026-entropy</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2398/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>54510</start>
<end>54525</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Entropy Equilibrium Sampling: Tuning-Free Truncation for Language Model Decoding
%A Cai, Xiaodong
%A Lin, Hai
%A Zhan, Shaoxiong
%A Luo, Weiqi
%A Zheng, Hai-Tao
%A Kim, Hong-Gee
%A Zhou, XiuTeng
%A Ma, Chunxia
%A Yang, Yu
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F cai-etal-2026-entropy
%X Token sampling strategies critically influence text generation quality in large language models (LLMs). However, existing methods introduce additional hyperparameters, requiring extensive tuning and complicating deployment. We present Entropy Equilibrium Sampling (EES), a tuning-free approach grounded in information theory that dynamically determines the candidate set size using a first-crossing rule: it stops candidate-set expansion at the first index where cumulative probability mass exceeds the normalized entropy of the top-k distribution and retains the preceding prefix. We evaluate EES on reasoning, mathematical, and open-ended generation tasks across multiple model architectures and scales. Our results show that EES achieves robust performance across temperature settings without retuning truncation hyperparameters, delivering competitive accuracy with tuned baselines while maintaining diversity. Analysis reveals that the retained candidate-set size k* automatically encodes temperature effects, adapting from near-greedy behavior at low temperatures to broad candidate retention at high temperatures. By eliminating truncation-specific hyperparameter tuning, EES simplifies deployment while providing stable and competitive performance. Our code is publicly available.
%U https://aclanthology.org/2026.emnlp-main.2398/
%P 54510-54525
Markdown (Informal)
[Entropy Equilibrium Sampling: Tuning-Free Truncation for Language Model Decoding](https://aclanthology.org/2026.emnlp-main.2398/) (Cai et al., EMNLP 2026)
ACL
- Xiaodong Cai, Hai Lin, Shaoxiong Zhan, Weiqi Luo, Hai-Tao Zheng, Hong-Gee Kim, XiuTeng Zhou, Chunxia Ma, and Yu Yang. 2026. Entropy Equilibrium Sampling: Tuning-Free Truncation for Language Model Decoding. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 54510–54525, Budapest, Hungary. Association for Computational Linguistics.