AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

DSB: Dynamic Sliding Block Scheduling for Diffusion LLMs

Lizhuo Luo, Shenggui Li, Yonggang Wen, Tianwei Zhang

ICML 2026 · Large language models › Language, safety and interpretability · approach: Parallel and early-exit decoding · 0 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
add DSB Cache prefix-window refresh instead of Dual Cache (ablation)DSB (const.) · GSM8K, LLaDA-8B-Instructaccuracy: 76.42 -> 80.14Improved
add DSB Cache prefix-window refresh instead of Dual Cache (ablation)DSB (const.) · GSM8K, LLaDA-8B-InstructTPS: 78.93 -> 98.10Improved
add DSB Cache prefix-window refresh instead of Dual Cache (ablation)DSB (const.) · HumanEval, LLaDA-8B-Instructaccuracy: 28.05 -> 37.80Improved
add DSB Cache prefix-window refresh instead of Dual Cache (ablation)DSB (greedy) · HumanEval, LLaDA-8B-Instructaccuracy: 28.05 -> 39.63Improved
add DSB Cache prefix-window refresh instead of Dual Cache (ablation)DSB (greedy) · GSM8K, LLaDA-8B-Instructaccuracy: 76.35 -> 80.29Improved
use DSB (const.) instead of AdaBlockconfidence decoding · HumanEval, LLaDA-8B-Instructaccuracy: 39.63 -> 42.07Improved
use DSB (greedy) instead of naive blockconfidence decoding · GSM8K, LLaDA-8B-Instructaccuracy: 77.26 -> 78.54Improved
use DSB (greedy) instead of naive blockconfidence decoding · GSM8K, LLaDA-8B-InstructTPS: 48.70 -> 51.03Improved
use DSB with DSB Cache instead of naive block with Dual Cacheconfidence decoding with cache · GSM8K, LLaDA-8B-Instructaccuracy: 77.40 -> 80.29Improved
use DSB with DSB Cache instead of naive block with Dual Cacheconfidence decoding with cache · GSM8K, LLaDA-8B-InstructTPS: 92.26 -> 99.61Improved
use larger initial block length of 64 for DSB (ablation)DSB · GSM8K, LLaDA-8B-InstructaccuracyHurt
increase max block length (ablation)DSB · LLaDA modelsaccuracyMixed

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