Denoising Time Matters:Diverse Generation in Diffusion Language Models
cs.CL, cs.AI
Submitted: 2026-01-30
Updated: 2026-09-26
Project page: https://taps-dlm.github.io
License: http://creativecommons.org/licenses/by/4.0/
The gist: Diffusion language models (Diffusion-LMs) generate text through iterative denoising, exposing a temporal structure that is largely absent from autoregressive decoding.
Terminology
Abstract
Diffusion language models (Diffusion-LMs) generate text through iterative denoising, exposing a temporal structure that is largely absent from autoregressive decoding. In this paper, we show that this temporal structure provides a useful control axis for generation diversity: early denoising steps mainly determine high-level semantic trajectories, while later steps refine lexical realization. Motivated by this observation, we propose Time-Annealed Perturbation Sampling (TAPS), a training-free inference strategy that samples nearby conditioning trajectories through time-aware, manifold-constrained perturbations. TAPS encourages semantic branching during early denoising and anneals the perturbation away before refinement, improving exploration while preserving prompt alignment, generation quality, and reasoning ability. Experiments on multiple Diffusion-LM backbones, including non-autoregressive and semi-autoregressive models, show that TAPS consistently improves semantic and lexical diversity across open-ended and instruction-following generation tasks, while preserving reasoning ability on verifiable reasoning benchmarks with negligible overhead.
Sources
- SDAR: A Synergistic Diffusion-AutoRegression Paradigm for Scalable Sequence Generation
- Large Language Diffusion Models
- Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models
- Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding
- Dream 7B: Diffusion Large Language Models
- A Survey on Diffusion Language Models
- TESS 2: A Large-Scale Generalist Diffusion Language Model
- Forcing Diffuse Distributions out of Language Models
- Preserving Diversity in Supervised Fine-Tuning of Large Language Models
- Diverse Preference Optimization
- Jointly Reinforcing Diversity and Quality in Language Model Generations
- NoveltyBench: Evaluating Language Models for Humanlike Diversity
- The Curious Case of Neural Text Degeneration
- Is Temperature the Creativity Parameter of Large Language Models?
- Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction
- Hierarchical Neural Story Generation
- Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researchers
- CADS: Unleashing the Diversity of Diffusion Models through Condition-Annealed Sampling
- Exploration-Driven Policy Optimization in RLHF: Theoretical Insights on Efficient Data Utilization
- SEED-GRPO: Semantic Entropy Enhanced GRPO for Uncertainty-Aware Policy Optimization
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering