AI Paper Atlas

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

Search or Accelerate: Confidence-Switched Position Beam Search for Diffusion Language Models

Mingyu Cao, Alvaro Correia, Christos Louizos, Shiwei Liu, Lu Yin

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
replace greedy with position beam search (single token)Dream-7B-Base · avg of HumanEval/MBPP/GSM8Kaccuracy: 60.1 -> 63.0 (speedup x0.54)Improved
replace greedy with position beam search (single token)LLaDA-8B-Base · avg of HumanEval/MBPP/GSM8Kaccuracy: 47.8 -> 48.9 (speedup x0.48)Improved
force 2 tokens per step in position beam search (ablation)LLaDA-8B-Base · avgaccuracy: 48.9 -> 44.9Hurt
force 2 tokens per step in position beam search (ablation)Dream-7B-Base · avgaccuracy: 63.0 -> 60.7Hurt
add confidence-switched search/parallel (SOAR) vs greedyDream-7B-Base · avgaccuracy: 60.1 -> 62.4 (speedup x1.43)Improved
add confidence-switched search/parallel (SOAR) vs greedyLLaDA-8B-Base · avgaccuracy: 47.8 -> 49.2 (speedup x1.62)Improved
raise confidence threshold tau from 0.8 to 0.98 (ablation)SOAR on Dream-7B · MBPP and GSM8Kaccuracy and speedupMixed
increase beam size from 1 to 4 (ablation)SOAR on Dream-7B · HumanEvalaccuracy and speedupMixed
use margin metric instead of max probability with SOARDream-7B-Base · HumanEval (256)pass@1: 47.0 -> 51.2Improved
use negative-entropy metric with SOARDream-7B-Base · HumanEval (512)pass@1: 56.1 -> 57.9Improved
apply SOAR to instruction-tuned model (pure diffusion)LLaDA-8B-Instruct · avgaccuracy: 49.1 -> 50.5Improved
apply SOAR to instruction-tuned model (block diffusion)LLaDA-8B-Instruct · HumanEval (512)pass@1: 48.8 -> 46.9Hurt

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