@inproceedings{tang-etal-2025-top,
title = "Top-$n\sigma$: Eliminating Noise in Logit Space for Robust Token Sampling of {LLM}",
author = "Tang, Chenxia and
Liu, Jianchun and
Xu, Hongli and
Huang, Liusheng",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.528/",
doi = "10.18653/v1/2025.acl-long.528",
pages = "10758--10774",
ISBN = "979-8-89176-251-0",
abstract = "Large language models (LLMs) rely heavily on sampling methods to generate diverse and high-quality text.While existing sampling methods like top-$p$ and min-$p$ have identified the detrimental effects of low-probability tails in LLMs' outputs, they still fail to effectively distinguish between diversity and noise. This limitation stems from their reliance on probability-based metrics that are inherently sensitive to temperature scaling. Through empirical and theoretical analysis, we make two key discoveries: (1) the pre-softmax logits exhibit a clear statistical separation between informative tokens and noise, and (2) we prove the mathematical equivalence of min-$p$ and top-(1-$p$) under uniform distribution over logits. These findings motivate the design of top-n$\sigma$, a novel sampling method that identifies informative tokens by eliminating noise directly in logit space.Unlike existing methods that become unstable at high temperatures, top-$n\sigma$ achieves temperature-invariant token selection while preserving output diversity. Extensive experiments across reasoning and creative writing tasks demonstrate that our method consistently outperforms existing approaches, with particularly significant improvements in high-temperature settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="tang-etal-2025-top">
<titleInfo>
<title>Top-nσ: Eliminating Noise in Logit Space for Robust Token Sampling of LLM</title>
</titleInfo>
<name type="personal">
<namePart type="given">Chenxia</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianchun</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hongli</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liusheng</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>Large language models (LLMs) rely heavily on sampling methods to generate diverse and high-quality text.While existing sampling methods like top-p and min-p have identified the detrimental effects of low-probability tails in LLMs’ outputs, they still fail to effectively distinguish between diversity and noise. This limitation stems from their reliance on probability-based metrics that are inherently sensitive to temperature scaling. Through empirical and theoretical analysis, we make two key discoveries: (1) the pre-softmax logits exhibit a clear statistical separation between informative tokens and noise, and (2) we prove the mathematical equivalence of min-p and top-(1-p) under uniform distribution over logits. These findings motivate the design of top-nσ, a novel sampling method that identifies informative tokens by eliminating noise directly in logit space.Unlike existing methods that become unstable at high temperatures, top-nσ achieves temperature-invariant token selection while preserving output diversity. Extensive experiments across reasoning and creative writing tasks demonstrate that our method consistently outperforms existing approaches, with particularly significant improvements in high-temperature settings.</abstract>
<identifier type="citekey">tang-etal-2025-top</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.528</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.528/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>10758</start>
<end>10774</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Top-nσ: Eliminating Noise in Logit Space for Robust Token Sampling of LLM
%A Tang, Chenxia
%A Liu, Jianchun
%A Xu, Hongli
%A Huang, Liusheng
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F tang-etal-2025-top
%X Large language models (LLMs) rely heavily on sampling methods to generate diverse and high-quality text.While existing sampling methods like top-p and min-p have identified the detrimental effects of low-probability tails in LLMs’ outputs, they still fail to effectively distinguish between diversity and noise. This limitation stems from their reliance on probability-based metrics that are inherently sensitive to temperature scaling. Through empirical and theoretical analysis, we make two key discoveries: (1) the pre-softmax logits exhibit a clear statistical separation between informative tokens and noise, and (2) we prove the mathematical equivalence of min-p and top-(1-p) under uniform distribution over logits. These findings motivate the design of top-nσ, a novel sampling method that identifies informative tokens by eliminating noise directly in logit space.Unlike existing methods that become unstable at high temperatures, top-nσ achieves temperature-invariant token selection while preserving output diversity. Extensive experiments across reasoning and creative writing tasks demonstrate that our method consistently outperforms existing approaches, with particularly significant improvements in high-temperature settings.
%R 10.18653/v1/2025.acl-long.528
%U https://aclanthology.org/2025.acl-long.528/
%U https://doi.org/10.18653/v1/2025.acl-long.528
%P 10758-10774
Markdown (Informal)
[Top-n𝜎: Eliminating Noise in Logit Space for Robust Token Sampling of LLM](https://aclanthology.org/2025.acl-long.528/) (Tang et al., ACL 2025)
ACL