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

2026-01-30

AI Summary

Diffusion language models struggle to effectively control semantic diversity during generation, limiting their capacity for multi-path reasoning and creative expression. This work reveals, for the first time, a temporal division of labor in their denoising process: early steps predominantly shape global semantics, while later steps refine local details. Building on this insight, the authors propose Temporal Annealing Perturbation Sampling (TAPS), a training-free method that injects controllable perturbations during early denoising stages to stimulate diverse semantic branches and gradually anneals these perturbations in later stages to preserve fluency and instruction adherence. TAPS is compatible with both non-autoregressive and semi-autoregressive architectures and significantly enhances output diversity across multiple creative writing and reasoning benchmarks without compromising generation quality.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

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.

Problem

• diffusion language models
• generation diversity
• temporal structure
• semantic branching
• text generation

Innovation

• Time-Annealed Perturbation Sampling
• Diffusion Language Models
• Generation Diversity
• Temporal Division of Labor
• Training-Free Inference