mlsanityData to 7 Oct 2026

AdaBlock-dLLM: Semantic-Aware Diffusion LLM Inference via Adaptive Block Size

arXiv 2509.26432 ↗First posted 30 Sep 2025cs.LG cs.AIcited by 30

On the map: Efficient LLM inference and serving → Diffusion LLM and MoE inference

Abstract (arXiv)

Diffusion-based large language models (dLLMs) are gaining attention for their inherent capacity for parallel decoding, offering a compelling alternative to autoregressive LLMs. Among various decoding strategies, block-wise semi-autoregressive (semi-AR) approaches are widely adopted due to their support for KV caching and their favorable accuracy-speed trade-off. However, this paper identifies two fundamental limitations in the conventional semi-AR decoding approach that applies a fixed block size: i) late decoding overhead, where the unmasking of high-confidence tokens outside the current block is unnecessarily delayed, and ii) premature decoding error, where low-confidence tokens inside the current block are committed too early, leading to incorrect tokens. This paper presents the first systematic investigation challenging the fixed block size setting in semi-AR decoding. Through a statistical analysis of confidence dynamics during the denoising process, we identify a volatility band (VB) region during dLLM decoding, which encodes local semantic structure and can be used to guide adaptive block sizing. Leveraging these insights, we introduce AdaBlock-dLLM, a training-free, plug-and-play scheduler that adaptively aligns block boundaries with semantic steps by adjusting block size during runtime. Extensive experiments across diverse benchmarks show that AdaBlock-dLLM achieves up to 5.3% accuracy improvement under the same throughput budget. Beyond inference-time optimization, we hope our semantics-aware adaptive scheduling approach and confidence-based analysis will inspire future training strategies for dLLMs. Our code is available at https://github.com/lgxi24/AdaBlock-dLLM.

Key facts. Cited by 30 papers in the map (data through 7 Oct 2026). Placed in Diffusion LLM and MoE inference, one of 12 directions in Efficient LLM inference and serving. Builds on 40 of its 45 references that are in the map. First posted on arXiv 30 Sep 2025.

Builds on

40 of 45 references.

Closest ancestors · 6

Same direction, newest first

  1. 2025Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing8 Aug
  2. 2025Sparse-dLLM: Accelerating Diffusion LLMs with Dynamic Cache Eviction4 Aug
  3. 2025Accelerating Diffusion LLMs via Adaptive Parallel Decoding31 May
  4. 2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding28 May
  5. 2025Dimple: Discrete Diffusion Multimodal Large Language Model with Parallel Decoding22 May
  6. 2025dKV-Cache: The Cache for Diffusion Language Models21 May

Other references · 34

  1. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 8,964
  4. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  5. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  6. 2022Scalable Diffusion Models with Transformerscited by 5,931
  7. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  8. 2021Program Synthesis with Large Language Modelscited by 3,509
  9. 2022Elucidating the Design Space of Diffusion-Based Generative Modelscited by 2,994
  10. 2022Video Diffusion Modelscited by 2,280
  11. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,928
  12. 2021Structured Denoising Diffusion Models in Discrete State-Spacescited by 1,665
22 more
  1. 2023Instruction-Following Evaluation for Large Language Modelscited by 1,143
  2. 2025Large Language Diffusion Modelscited by 872
  3. 2023Discrete Diffusion Modeling by Estimating the Ratios of the Data Distributioncited by 712
  4. 2021Argmax Flows and Multinomial Diffusion: Learning Categorical Distributionscited by 566
  5. 2025Dream 7B: Diffusion Large Language Modelscited by 515
  6. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Modelscited by 385
  7. 2025LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Modelscited by 273
  8. 2025MMaDA: Multimodal Large Diffusion Language Modelscited by 241
  9. 2025DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generationcited by 184
  10. 2025Remasking Discrete Diffusion Models with Inference-Time Scalingcited by 181
  11. 2025Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inferencecited by 173
  12. 2025dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Cachingcited by 168
  13. 2025LaViDa: A Large Diffusion Language Model for Multimodal Understandingcited by 81
  14. 2025FlashDLM: Accelerating Diffusion Language Model Inference via Efficient KV Caching and Guided Diffusioncited by 77
  15. 2025Diffusion Language Models Know the Answer Before Decodingcited by 68
  16. 2025Diffusion Beats Autoregressive in Data-Constrained Settingscited by 61
  17. 2025Wide-In, Narrow-Out: Revokable Decoding for Efficient and Effective DLLMscited by 45
  18. 2025DPad: Efficient Diffusion Language Models with Suffix Dropoutcited by 38
  19. 2025Plan for Speed: Dilated Scheduling for Masked Diffusion Language Modelscited by 34
  20. 2025Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Modelscited by 33
  21. 2025MDPO: Overcoming the Training-Inference Divide of Masked Diffusion Language Modelscited by 32
  22. 2025A Survey on Parallel Text Generation: From Parallel Decoding to Diffusion Language Modelscited by 27

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 4; 2026: 26.

30 papers.

Sort:

By area:

  1. 2025Attention Is All You Need for KV Cache in Diffusion LLMscited by 32
  2. 2026LocateAnything: Fast and High-Quality Vision-Language Grounding with Parallel Box Decodingcited by 20
  3. 2026Deferred Commitment Decoding for Diffusion Language Modelscited by 11
  4. 2025Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibrationcited by 7
  5. 2026DSB: Dynamic Sliding Block Scheduling for Diffusion LLMscited by 7
  6. 2026When to Commit? Towards Variable-Size Self-Contained Blocks for Discrete Diffusion Language Modelscited by 7
  7. 2025Taming the Memory Footprint Crisis: System Design for Production Diffusion LLM Servingcited by 5
  8. 2026Dynamic Expert Sharing: Decoupling Memory from Parallelism in Mixture-of-Experts Diffusion LLMscited by 4
  9. 2026Swordsman: Entropy-Driven Adaptive Block Partition for Efficient Diffusion Language Modelscited by 3
  10. 2026Breaking Block Boundaries: Anchor-based History-stable Decoding for Diffusion Large Language Modelscited by 3
  11. 2026Multi-Block Diffusion Language Modelscited by 2
  12. 2026VDLM: Variable Diffusion LMs via Robust Latent-to-Text Renderingcited by 1
18 more
  1. 2026DepCap: Adaptive Block-Wise Parallel Decoding for Efficient Diffusion LM Inferencecited by 1
  2. 2026Break the Block: Dynamic-size Reasoning Blocks for Diffusion Large Language Models via Monotonic Entropy Descent with Reinforcement Learningcited by 1
  3. 2026Adaptive Block Diffusion: Resolving Training-Inference Mismatch in Diffusion Language Modelscited by 1
  4. 2026Polestar: Drift-Aware Cache Calibration and Token Commitment for Efficient Inference of Diffusion LLMscited by 1
  5. 2025STaRR: Spatial-Temporal Token-Dynamics-Aware Responsive Remasking for Diffusion Language Modelscited by 0
  6. 2026Dystruct: Dynamically Structured Diffusion Language Model Decoding via Bayesian Inferencecited by 0
  7. 2026Block-R1: Rethinking the Role of Block Size in Multi-domain Reinforcement Learning for Diffusion Large Language Modelscited by 0
  8. 2026Dynamic Chunking for Diffusion Language Modelscited by 0
  9. 2026When Confidence Misleads: Suffix Anchoring and Anchor-Proximity Confidence Modulation for Diffusion Language Modelscited by 0
  10. 2026SemBlock: Semantic Boundary Dynamic Blocks for Diffusion LLMscited by 0
  11. 2026AsyncLane: Decoupling Refinement from Advancement in Diffusion Language Model Decodingcited by 0
  12. 2026CORA-Diff: Confidence-Oriented Residual Acceptance for Efficient Diffusion Language Model Inferencecited by 0
  13. 2026Context-Aware Cluster Decoding: Semantic Anchor-Driven Coherence in dMLLMscited by 0
  14. 2026Affix Cache for Diffusion Large Language Modelscited by 0
  15. 2026Information-Guided Frontier Decoding: Contextual Utility-Driven Commitment in dMLLMscited by 0
  16. 2026Early-Bird Decoding: Accelerating Diffusion LLMs with Learnable Block Sizes and Parallel Samplingcited by 0
  17. 2026Optimizing Denoising Trajectories in dLLMs: A Lightweight Evolutionary Heuristic Approachcited by 0
  18. 2026Faster Block-Diffusion Serving with Distribution-Free Risk Guaranteescited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Fast-dLLM v2: Efficient Block-Diffusion LLM
  2. 2026Fast-dLLM++: Fr'echet Profile Decoding for Faster Diffusion LLM Inference
  3. 2026Dynamic-dLLM: Dynamic Cache-Budget and Adaptive Parallel Decoding for Training-Free Acceleration of Diffusion LLM
  4. 2026S2D2: Fast Decoding for Diffusion LLMs via Training-Free Self-Speculation
  5. 2026PACE-dLLM: Elastic Block Decoding via Confidence Cliff Estimation for Diffusion Language Models
  6. 2026d3LLM: Ultra-Fast Diffusion LLM using Pseudo-Trajectory Distillation
  7. 2025ParallelBench: Understanding the Trade-offs of Parallel Decoding in Diffusion LLMs
  8. 2026Streaming-dLLM: Accelerating Diffusion LLMs via Suffix Pruning and Dynamic Decoding