Time-Annealed Perturbation Sampling: Diverse Generation for Diffusion Language Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wu, Jingxuan, Wan, Zhenglin, Yu, Xingrui, Yang, Yuzhe, Huang, Yiqiao, Tsang, Ivor, You, Yang
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910055025082368
author Wu, Jingxuan
Wan, Zhenglin
Yu, Xingrui
Yang, Yuzhe
Huang, Yiqiao
Tsang, Ivor
You, Yang
author_facet Wu, Jingxuan
Wan, Zhenglin
Yu, Xingrui
Yang, Yuzhe
Huang, Yiqiao
Tsang, Ivor
You, Yang
contents Diffusion language models (Diffusion-LMs) introduce an explicit temporal dimension into text generation, yet how this structure can be leveraged to control generation diversity for exploring multiple valid semantic or reasoning paths remains underexplored. In this paper, we show that Diffusion-LMs, like diffusion models in image generation, exhibit a temporal division of labor: early denoising steps largely determine the global semantic structure, while later steps focus on local lexical refinement. Building on this insight, we propose Time-Annealed Perturbation Sampling (TAPS), a training-free inference strategy that encourages semantic branching early in the diffusion process while progressively reducing perturbations to preserve fluency and instruction adherence. TAPS is compatible with both non-autoregressive and semi-autoregressive Diffusion backbones, demonstrated on LLaDA and TraDo in our paper, and consistently improves output diversity across creative writing and reasoning benchmarks without compromising generation quality.
format Preprint
id arxiv_https___arxiv_org_abs_2601_22629
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Time-Annealed Perturbation Sampling: Diverse Generation for Diffusion Language Models
Wu, Jingxuan
Wan, Zhenglin
Yu, Xingrui
Yang, Yuzhe
Huang, Yiqiao
Tsang, Ivor
You, Yang
Computation and Language
Artificial Intelligence
Diffusion language models (Diffusion-LMs) introduce an explicit temporal dimension into text generation, yet how this structure can be leveraged to control generation diversity for exploring multiple valid semantic or reasoning paths remains underexplored. In this paper, we show that Diffusion-LMs, like diffusion models in image generation, exhibit a temporal division of labor: early denoising steps largely determine the global semantic structure, while later steps focus on local lexical refinement. Building on this insight, we propose Time-Annealed Perturbation Sampling (TAPS), a training-free inference strategy that encourages semantic branching early in the diffusion process while progressively reducing perturbations to preserve fluency and instruction adherence. TAPS is compatible with both non-autoregressive and semi-autoregressive Diffusion backbones, demonstrated on LLaDA and TraDo in our paper, and consistently improves output diversity across creative writing and reasoning benchmarks without compromising generation quality.
title Time-Annealed Perturbation Sampling: Diverse Generation for Diffusion Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2601.22629