AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Stopping Computation for Converged Tokens in Masked Diffusion-LM Decoding

Daisuke Oba, Danushka Bollegala, Masahiro Kaneko, Naoaki Okazaki

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
lock converged token positions (SureLock) in decodingLLaDA-8B-Instruct · MT-Bench subset, Ngen=256algorithmic FLOPs ratio: 1.0 -> 0.496Improved
lock converged positions (SureLock)LLaDA-8B-Instruct · MT-Bench single-turnGPT-4o judge score: 4.8 -> 4.9 (Ngen=128)No clear effect
lock converged positions (SureLock)LLaDA-8B-Base · WikiText-103, Ngen=64, steps=64Gen.-PPL: 3.4813 -> 4.5722Hurt
lock converged positions (SureLock)LLaDA-8B-Base · WikiText-103, Ngen=512, steps=512Gen.-PPL: 1.4658 -> 1.5248No clear effect
tighten KL threshold from 5e-3 to 5e-8 (ablation)SureLock · WikiText-103, Ngen=64Gen.-PPL: 4.7596 -> 4.0033Improved
tighten KL threshold from 5e-2 to 5e-4 (ablation)SureLock · MT-Bench subset, Ngen=64FLOPs ratio: 0.482 -> 0.547Hurt
combine SureLock with selection-based sparsification (ablation)SureLock · MT-Bench, Ngen=256FLOPs ratio: 0.54 -> 0.28Improved
apply SureLock on HumanEval code generationLLaDA-8B-Instruct · HumanEval (first 70)Pass@1: 18.6 -> 18.6-20.3No clear effect
apply SureLock at small workloadLLaDA-8B-Instruct · runtime, Ngen=64, B=1tokens per secondNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy