mlsanityData to 14 Oct 2026

Stopping Computation for Converged Tokens in Masked Diffusion-LM Decoding

arXiv 2602.06412 ↗First posted 6 Feb 2026cs.CL cs.LGcited by 3

On the map: Diffusion models → Discrete and text diffusion

Abstract (arXiv)

Masked Diffusion Language Models generate sequences via iterative sampling that progressively unmasks tokens. However, they still recompute the attention and feed-forward blocks for every token position at every step -- even when many unmasked tokens are essentially fixed, resulting in substantial waste in compute. We propose SureLock: when the posterior at an unmasked position has stabilized across steps (our sure condition), we lock that position -- thereafter skipping its query projection and feed-forward sublayers -- while caching its attention keys and values so other positions can continue to attend to it. This reduces the dominant per-iteration computational cost from O(N2d) to O(MNd) where N is the sequence length, M is the number of unlocked token positions, and d is the model dimension. In practice, M decreases as the iteration progresses, yielding substantial savings. On LLaDA-8B, SureLock reduces algorithmic FLOPs by 30--50% relative to the same sampler without locking, while maintaining comparable generation quality. We also provide a theoretical analysis to justify the design rationale of SureLock: monitoring only the local KL at the lock step suffices to bound the deviation in final token probabilities. Our project page is available at https://daioba.github.io/surelock .

Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 21 of its 22 references that are in the map. First posted on arXiv 6 Feb 2026.

Builds on

21 of 22 references.

Closest ancestors · 8

Same direction, newest first

  1. 2025Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models23 Jun
  2. 2025LongLLaDA: Unlocking Long Context Capabilities in Diffusion LLMs17 Jun
  3. 2025dKV-Cache: The Cache for Diffusion Language Models21 May
  4. 2025CtrlDiff: Boosting Large Diffusion Language Models with Dynamic Block Prediction and Controllable Generation20 May
  5. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models12 Mar
  6. 2025Large Language Diffusion Models14 Feb
  7. 2024Simple and Effective Masked Diffusion Language Models11 Jun
  8. 2024Simplified and Generalized Masked Diffusion for Discrete Data6 Jun

Other references · 13

  1. 2017Attention Is All You Needcited by 60,361
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  4. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  5. 2016Pointer Sentinel Mixture Modelscited by 3,096
  6. 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Stepscited by 1,979
  7. 2022Diffusion-LM Improves Controllable Text Generationcited by 1,027
  8. 2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decodingcited by 420
  9. 2019Levenshtein Transformercited by 199
  10. 2025dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Cachingcited by 169
  11. 2024Simple Guidance Mechanisms for Discrete Diffusion Modelscited by 155
  12. 2025Accelerating Diffusion LLMs via Adaptive Parallel Decodingcited by 112
1 more
  1. 2025Shifting Long-Context LLMs Research from Input to Outputcited by 14

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Just on Time: Token-Level Early Stopping for Diffusion Language Modelscited by 2
  2. 2026Drifting Objectives for Refining Discrete Diffusion Language Modelscited by 2
  3. 2026What does FFN compression change downstream? Same-state causal restoration in diffusion language modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Stop the Flip-Flop: Context-Preserving Verification for Fast Revocable Diffusion Decoding
  2. 2025Lookahead Unmasking Elicits Accurate Decoding in Diffusion Language Models
  3. 2026Mask-Guided KV Cache Eviction in Block Diffusion Language Models
  4. 2026One Block Ahead: Promoting Sampler State in Diffusion Language Models
  5. 2026Supportive Token Revealing for Fast Diffusion Language Model Decoding
  6. 2026Low-Confidence Remasking Traps Flexibility: Realizing Arbitrary-Order Potential for Diverse Rollouts in Diffusion LLMs
  7. 2026x-Prediction Flow: Efficient Continuous Decoding for Masked Diffusion Language Models
  8. 2026NAVIRA: Decoupled Stochastic Remasking for Masked Diffusion Language Models