# Stopping Computation for Converged Tokens in Masked Diffusion-LM Decoding

arXiv: https://arxiv.org/abs/2602.06412 · HTML: https://mlsanity.com/paper/2602.06412/

## 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.

## Placed in

- Diffusion models / [Discrete and text diffusion](https://mlsanity.com/diffusion/discrete-and-text-diffusion/)

## Builds on (most cited first, 21 of 21)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Pointer Sentinel Mixture Models](https://mlsanity.com/paper/1609.07843/) (2016)
- [DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps](https://mlsanity.com/paper/2206.00927/) (2022)
- [Diffusion-LM Improves Controllable Text Generation](https://mlsanity.com/paper/2205.14217/) (2022)
- [Large Language Diffusion Models](https://mlsanity.com/paper/2502.09992/) (2025)
- [Simple and Effective Masked Diffusion Language Models](https://mlsanity.com/paper/2406.07524/) (2024)
- [Simplified and Generalized Masked Diffusion for Discrete Data](https://mlsanity.com/paper/2406.04329/) (2024)
- [Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding](https://mlsanity.com/paper/2505.22618/) (2025)
- [Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models](https://mlsanity.com/paper/2503.09573/) (2025)
- [Levenshtein Transformer](https://mlsanity.com/paper/1905.11006/) (2019)
- [dLLM-Cache: Accelerating Diffusion Large Language Models with Adaptive Caching](https://mlsanity.com/paper/2506.06295/) (2025)
- [Simple Guidance Mechanisms for Discrete Diffusion Models](https://mlsanity.com/paper/2412.10193/) (2024)
- [dKV-Cache: The Cache for Diffusion Language Models](https://mlsanity.com/paper/2505.15781/) (2025)
- [Accelerating Diffusion LLMs via Adaptive Parallel Decoding](https://mlsanity.com/paper/2506.00413/) (2025)
- [LongLLaDA: Unlocking Long Context Capabilities in Diffusion LLMs](https://mlsanity.com/paper/2506.14429/) (2025)
- [Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models](https://mlsanity.com/paper/2506.19037/) (2025)
- [CtrlDiff: Boosting Large Diffusion Language Models with Dynamic Block Prediction and Controllable Generation](https://mlsanity.com/paper/2505.14455/) (2025)
- [Shifting Long-Context LLMs Research from Input to Output](https://mlsanity.com/paper/2503.04723/) (2025)

## Built on by (most cited first, 3 of 3)

- [Just on Time: Token-Level Early Stopping for Diffusion Language Models](https://mlsanity.com/paper/2602.11133/) (2026)
- [Drifting Objectives for Refining Discrete Diffusion Language Models](https://mlsanity.com/paper/2605.19470/) (2026)
- [What does FFN compression change downstream? Same-state causal restoration in diffusion language models](https://mlsanity.com/paper/2609.31685/) (2026)

## 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 .