mlsanityData to 7 Oct 2026

Progressive Refinement Regulation for Accelerating Diffusion Language Model Decoding

arXiv 2603.04514 ↗First posted 4 Mar 2026cs.AIcited by 0

On the map: Diffusion models → Discrete and text diffusion

Abstract (arXiv)

Diffusion language models generate text through iterative denoising under a uniform refinement rule applied to all tokens. However, tokens stabilize at different rates in practice, leading to substantial redundant refinement and motivating refinement control over the denoising process. Existing approaches typically assess refinement necessity from instantaneous, step-level signals under a fixed decoding process. In contrast, whether a token has converged is defined by how its prediction changes along its future refinement trajectory. Moreover, changing the refinement rule reshapes future refinement trajectories, which in turn determine how refinement rules should be formulated, making refinement control inherently dynamic. We propose Progressive Refinement Regulation (PRR), a progressive, trajectory-grounded refinement control framework that derives a token-level notion of empirical convergence progress from full decoding rollouts. Based on this signal, PRR learns a lightweight token-wise controller to regulate refinement via temperature-based distribution shaping under a progressive self-evolving training scheme. Experiments show that PRR substantially accelerates diffusion language model decoding while preserving generation quality.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 62 of its 74 references that are in the map. First posted on arXiv 4 Mar 2026.

Builds on

62 of 74 references.

Closest ancestors · 31

Same direction, newest first

  1. 2025Learning Unmasking Policies for Diffusion Language Models9 Dec
  2. 2025Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules2 Dec
  3. 2025SPG: Sandwiched Policy Gradient for Masked Diffusion Language Models10 Oct
  4. 2025Guided Star-Shaped Masked Diffusion9 Oct
  5. 2025Improving Discrete Diffusion Unmasking Policies Beyond Explicit Reference Policies7 Oct
  6. 2025Taming Masked Diffusion Language Models via Consistency Trajectory Reinforcement Learning with Fewer Decoding Step28 Sep
  7. 2025Diffusion Language Models Know the Answer Before Decoding27 Aug
  8. 2025Dream 7B: Diffusion Large Language Models21 Aug
  9. 2025DPad: Efficient Diffusion Language Models with Suffix Dropout19 Aug
  10. 2025Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Models12 Aug
  11. 2025A Survey on Parallel Text Generation: From Parallel Decoding to Diffusion Language Models12 Aug
  12. 2025Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inference4 Aug
19 more
  1. 2025DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generation25 Jun
  2. 2025Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models23 Jun
  3. 2025Accelerated Sampling from Masked Diffusion Models via Entropy Bounded Unmasking30 May
  4. 2025dKV-Cache: The Cache for Diffusion Language Models21 May
  5. 2025CtrlDiff: Boosting Large Diffusion Language Models with Dynamic Block Prediction and Controllable Generation20 May
  6. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models12 Mar
  7. 2025Remasking Discrete Diffusion Models with Inference-Time Scaling1 Mar
  8. 2025Large Language Diffusion Models14 Feb
  9. 2025Train for the Worst, Plan for the Best: Understanding Token Ordering in Masked Diffusions10 Feb
  10. 2025Accelerated Diffusion Models via Speculative Sampling9 Jan
  11. 2024Beyond Autoregression: Fast LLMs via Self-Distillation Through Time28 Oct
  12. 2024Scaling up Masked Diffusion Models on Text24 Oct
  13. 2024Beyond Autoregression: Discrete Diffusion for Complex Reasoning and Planning18 Oct
  14. 2024Think While You Generate: Discrete Diffusion with Planned Denoising8 Oct
  15. 2024Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Sampling4 Sep
  16. 2024Simple and Effective Masked Diffusion Language Models11 Jun
  17. 2024Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data6 Jun
  18. 2023A Reparameterized Discrete Diffusion Model for Text Generation11 Feb
  19. 2021Autoregressive Diffusion Models5 Oct

Other references · 31

  1. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  2. 2017Proximal Policy Optimization Algorithmscited by 14,441
  3. 2015Distilling the Knowledge in a Neural Networkcited by 10,076
  4. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  5. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  6. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  7. 2016Layer Normalizationcited by 5,579
  8. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  9. 2017On Calibration of Modern Neural Networkscited by 4,860
  10. 2021Program Synthesis with Large Language Modelscited by 3,509
  11. 2016Gaussian Error Linear Units (GELUs)cited by 3,431
  12. 2023A Survey of Large Language Modelscited by 2,537
19 more
  1. 2015Trust Region Policy Optimizationcited by 2,503
  2. 2023Consistency Modelscited by 1,778
  3. 2021Structured Denoising Diffusion Models in Discrete State-Spacescited by 1,665
  4. 2023Instruction-Following Evaluation for Large Language Modelscited by 1,143
  5. 2022MaskGIT: Masked Generative Image Transformercited by 1,082
  6. 2023Accelerating Large Language Model Decoding with Speculative Samplingcited by 975
  7. 2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decodingcited by 404
  8. 2024Discrete Flow Matchingcited by 301
  9. 2025Mercury: Ultra-Fast Language Models Based on Diffusioncited by 225
  10. 2025dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Cachingcited by 168
  11. 2025Accelerating Diffusion LLMs via Adaptive Parallel Decodingcited by 108
  12. 2024Speculative Diffusion Decoding: Accelerating Language Generation through Diffusioncited by 57
  13. 2023A Comprehensive Survey on Knowledge Distillation of Diffusion Modelscited by 45
  14. 2024textitJump Your Steps: Optimizing Sampling Schedule of Discrete Diffusion Modelscited by 40
  15. 2025Structuring The Future: Diffusion LLM Speculative Decoding via Calibrated Draft Graphscited by 21
  16. 2024A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Modelscited by 13
  17. 2025Thinking Inside the Mask: In-Place Prompting in Diffusion LLMscited by 13
  18. 2025Latent Refinement Decoding: Enhancing Diffusion-Based Language Models by Refining Belief Statescited by 9
  19. 2025Dimathtt[M]O: Distilling Masked Diffusion Models into One-step Generatorcited by 5

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Remask, Don't Replace: Token-to-Mask Refinement in Diffusion Language Models
  2. 2026AsyncLane: Decoupling Refinement from Advancement in Diffusion Language Model Decoding
  3. 2026CoDR: Training-Free Confidence-Drift Remasking for Diffusion Language Models
  4. 2026Forward-Free Diffusion Language Models with BPTT-Free Looped Refinement
  5. 2026Efficient Diffusion LLMs via Temporal-Spatial Parallel Decoding and Confidence Extrapolation
  6. 2026Just on Time: Token-Level Early Stopping for Diffusion Language Models
  7. 2026Edit-Based Refinement for Parallel Masked Diffusion Language Models
  8. 2026CORA-Diff: Confidence-Oriented Residual Acceptance for Efficient Diffusion Language Model Inference