AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Beyond Fixed: Training-Free Variable-Length Denoising for Diffusion Large Language Models

Jinsong Li, Xiaoyi Dong, Yuhang Zang, Yuhang Cao, Jiaqi Wang, Dahua Lin

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 48 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
use DAEDAL instead of best fixed lengthLLaDA-Instruct-8B · MATH500accuracy: 39.6 -> 44.2Improved
use DAEDAL instead of best fixed lengthLLaDA-Instruct-8B · GSM8Kaccuracy: 83.8 -> 85.8Improved
use DAEDAL instead of best fixed lengthLLaDA-1.5-8B · MATH500accuracy: 43.6 -> 42.4Hurt
use DAEDAL instead of best fixed lengthLLaDA-1.5-8B · HumanEvalaccuracy: 50.0 -> 48.8Hurt
use DAEDAL instead of best fixed lengthLLaDA-1.5-8B · Average of 4 benchmarksaccuracy: 54.15 -> 54.23No clear effect
add DAEDAL at same initial length 64LLaDA-Instruct-8B · GSM8Kaccuracy: 48.0 -> 85.8Improved
add DAEDAL for efficiency (lower total tokens)LLaDA-Instruct-8B · GSM8Keffective token ratio: 27.7% -> 73.5%Improved
add Initial Length Adjustment (Stage 1) to Stage 2 (ablation)DAEDAL · GSM8Kaccuracy: 72.3 -> 85.8Improved
add Iterative Mask Insertion (Stage 2) to Stage 1 (ablation)DAEDAL · GSM8Kaccuracy: 84.1 -> 85.8Improved
raise initial length for Stage 2 only (ablation)DAEDAL Stage 2 · GSM8Kaccuracy: 72.3 -> 84.7Improved
increase initial length from 64 to 512 (ablation)DAEDAL · GSM8Kaccuracy: 85.8 -> 85.1No clear effect
increase expansion factor from 8 to 24 (ablation)DAEDAL · GSM8Kaccuracy: 85.8 -> 86.4No 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