Saved in:
Bibliographic Details
Main Authors: He, Siyang, Wang, Qiqi, Liu, Xiaoran, Ma, Hongnan, Shi, Yiwei, Song, Yuerong, Zhu, Ying, Liang, Tianyi, Huang, Zengfeng, He, Ziwei, Qiu, Xipeng
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.23182
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910006297755648
author He, Siyang
Wang, Qiqi
Liu, Xiaoran
Ma, Hongnan
Shi, Yiwei
Song, Yuerong
Zhu, Ying
Liang, Tianyi
Huang, Zengfeng
He, Ziwei
Qiu, Xipeng
author_facet He, Siyang
Wang, Qiqi
Liu, Xiaoran
Ma, Hongnan
Shi, Yiwei
Song, Yuerong
Zhu, Ying
Liang, Tianyi
Huang, Zengfeng
He, Ziwei
Qiu, Xipeng
contents Despite the non-autoregressive potential of diffusion language models (dLLMs), existing decoding strategies demonstrate positional bias, failing to fully unlock the potential of arbitrary generation. In this work, we delve into the inherent spectral characteristics of dLLMs and present the first frequency-domain analysis showing that low-frequency components in hidden states primarily encode global structural information and long-range dependencies, while high-frequency components are responsible for characterizing local details. Based on this observation, we propose FourierSampler, which leverages a frequency-domain sliding window mechanism to dynamically guide the model to achieve a "structure-to-detail" generation. FourierSampler outperforms other inference enhancement strategies on LLADA and SDAR, achieving relative improvements of 20.4% on LLaDA1.5-8B and 16.0% on LLaDA-8B-Instruct. It notably surpasses similarly sized autoregressive models like Llama3.1-8B-Instruct.
format Preprint
id arxiv_https___arxiv_org_abs_2601_23182
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FourierSampler: Unlocking Non-Autoregressive Potential in Diffusion Language Models via Frequency-Guided Generation
He, Siyang
Wang, Qiqi
Liu, Xiaoran
Ma, Hongnan
Shi, Yiwei
Song, Yuerong
Zhu, Ying
Liang, Tianyi
Huang, Zengfeng
He, Ziwei
Qiu, Xipeng
Computation and Language
Despite the non-autoregressive potential of diffusion language models (dLLMs), existing decoding strategies demonstrate positional bias, failing to fully unlock the potential of arbitrary generation. In this work, we delve into the inherent spectral characteristics of dLLMs and present the first frequency-domain analysis showing that low-frequency components in hidden states primarily encode global structural information and long-range dependencies, while high-frequency components are responsible for characterizing local details. Based on this observation, we propose FourierSampler, which leverages a frequency-domain sliding window mechanism to dynamically guide the model to achieve a "structure-to-detail" generation. FourierSampler outperforms other inference enhancement strategies on LLADA and SDAR, achieving relative improvements of 20.4% on LLaDA1.5-8B and 16.0% on LLaDA-8B-Instruct. It notably surpasses similarly sized autoregressive models like Llama3.1-8B-Instruct.
title FourierSampler: Unlocking Non-Autoregressive Potential in Diffusion Language Models via Frequency-Guided Generation
topic Computation and Language
url https://arxiv.org/abs/2601.23182