AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

LUGS: Latent-aware Guidance for Efficient Unmasking in Diffusion Large Language Models

Nuanqiao Shan, Kairong Han, Xinpeng Dong, Kun Kuang

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers

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 latent-aware guidance to beam searchBeam Search, LLaDA-8B-Instruct · HumanEvalPass@1: 35.37 -> 42.68Improved
add latent-aware guidance to beam searchBeam Search, LLaDA-8B-Instruct · Countdownaccuracy: 23.14 -> 26.07Improved
add latent-aware guidance to beam searchBeam Search, LLaDA-8B-Instruct · GSM8Kaccuracy: 79.38 -> 79.38No clear effect
add latent-aware guidance to beam searchBeam Search, LLaDA-1.5 · code tasks (HumanEval, MBPP)Pass@1: 40.02 -> 42.04Improved
add latent-aware guidance to beam searchBeam Search, LLaDA-1.5 · Math500accuracy: 35.40 -> 37.60Improved
add latent-aware guidance to beam searchBeam Search, LLaDA-1.5 · ARC-Challengeaccuracy: 84.04 -> 84.56No clear effect
use beam search instead of probability-based greedy unmaskingLLaDA-8B-Instruct · Countdownaccuracy: 19.34 -> 23.14Improved
pure latent score (beta=0.0) in selection score (ablation)LUGS · Countdownaccuracy: 23.14 -> 23.83No clear effect
vary guidance weight beta within 0.3 to 0.8 (ablation)LUGS · CountdownaccuracyNo clear effect
increase beam width from 2 to 9 (ablation)LUGS · Countdown, HumanEval, MBPPaccuracyNo clear effect
increase beam width (ablation)LUGS · inference timetime per sampleHurt
add LUGS guidance across beam widths (ablation)Beam Search · HumanEvalPass@1: average improvement 3.51%Improved

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