AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Consistent Diffusion Language Models

Hasan Amin, Yuan Gao, Yaser Souri, Subhojit Som, Ming Yin, Rajiv Khanna, Xia Song

ICML 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model 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
train with multi-path discrete consistency instead of MDLM objective110M masked diffusion LM · OpenWebText unconditional, 8 stepsgenerative perplexity: 682.7 -> 246.9Improved
tune CDLM for perplexity (shrink max step range, anneal max-step weight) (ablation)MCDLM · OpenWebText unconditional, 64 stepsgenerative perplexity: 67.7 -> 40.1Improved
distill MCDLM with SDTT objectiveMCDLM · OpenWebText unconditional, 1024 stepsgenerative perplexity: 31.2 -> 27.1 vs SDTT on MDLMImproved
use uniform prior CDLM instead of UDLM/DUOUCDLM · OpenWebText unconditional, 1024 stepsgenerative perplexity: 84.0 -> 71.7Improved
remove max-step diffusion anchor (ablation)CDLM · OpenWebText unconditionalentropy: 5.3 -> 3.2 at 8 stepsHurt
raise max-step weight from 0.4 to 1.0 (ablation)CDLM · OpenWebText unconditional, 64 stepsgenerative perplexity: 32.8 -> 75.1Hurt
use forward KL instead of JS divergence (ablation)CDLM · OpenWebText unconditional, 8 stepsgenerative perplexity: 110.6 -> 8e4Hurt
use backward KL instead of JS divergence (ablation)CDLM · OpenWebText unconditional, 1024 stepsgenerative perplexity: 19.7 -> 44.2 with entropy 5.2 -> 4.3Hurt
use staged increasing step-size scheduler instead of random (ablation)CDLM · OpenWebText unconditional, 8 stepsgenerative perplexity: 110.6 -> 160.8Hurt
use linear increasing step-size scheduler instead of random (ablation)CDLM · OpenWebText unconditionalgenerative perplexity: 19.7 -> 17.9 at 1024 steps; 110.6 -> 117.6 at 8 stepsMixed
use CDLM instead of DUO-DCD greedy for conditional generationMCDLM-PPLOptimized · Conditional generation (OOD sets)BLEU2: 26.7 -> 62.8 on OpenWebText 8 stepImproved

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