AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Soft-Masked Diffusion Language Models

Michael Hersche, Samuel Moor-Smith, Thomas Hofmann, Abbas Rahimi

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 17 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 soft-masking feedback (train from scratch)MDLM 169M, standard unmasking · OpenWebText, NFE 1/1MAUVE: 0.034 -> 0.602Improved
add soft-masking feedback (train from scratch)MDLM 169M, standard unmasking · OpenWebText, NFE 1/1generative perplexity: 50.46 -> 23.53Improved
add soft-masking on top of ReMDM remaskingMDLM 169M with ReMDM · OpenWebText, NFE 1/1MAUVE: 0.411 -> 0.774Improved
add soft-masking via 100k-step continued pretrainingpretrained binary MDLM · OpenWebText validationvalidation perplexity: 23.14 -> 21.63Improved
use iso-compute (half the update steps) instead of iso-update training for soft-masking (ablation)MDLM 169M SM · OpenWebText, NFE 1/4MAUVE: 0.383 -> 0.417No clear effect
add soft-masking finetuningDream-7B · MBPP+ at NFE 1/2accuracy (%): 39.6 -> 54.7Improved
add soft-masking finetuningDream-Coder-7B · MBPP at NFE 1/4accuracy (%): 25.9 -> 33.2Improved
add soft-masking finetuning at full NFE budgetDream-7B · HumanEval, HumanEval+, MBPP, MBPP+ at NFE 1/1accuracy (%): HumanEval 59.5 -> 57.8; HumanEval+ 53.0 -> 50.0Hurt
add soft-masking finetuning at full NFE budgetDream-Coder-7B · coding benchmarks at NFE 1/1accuracy (%): HumanEval 75.7 -> 76.2No clear effect
use top-k=1 versus k=3 superposition (ablation)soft-masking · coding vs language modelingaccuracy / perplexityMixed
use softmax with trainable temperature instead of top-k (ablation)soft-masking · language modeling and codingvalidation perplexity / accuracyNo clear effect
apply soft-masking only in early decoding steps (ablation)soft-masking · decoding scheduleperformanceImproved

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