AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Beyond Masks: Efficient, Flexible Diffusion Language Models via Deletion-Insertion Processes

Fangyu Ding, Ding Ding, Sijin Chen, Kaibo Wang, Peng Xu, Zijin Feng, Haoli Bai, Kai Han, Youliang Yan, Binhang Yuan, Jiacheng Sun

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 3 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 masking with deletion-insertion and train for FLOPs-aligned 800K steps (DID-F) vs RADDRADD small, OWT · zero-shot WikiTextperplexity: 38.27 -> 36.91Improved
replace masking with deletion-insertion, steps-aligned (DID-S) vs RADDRADD small, OWT · zero-shot PTBperplexity: 108.79 -> 115.37Hurt
DID-F vs RADD on medium modelsRADD medium, OWT · zero-shot PTBperplexity: 81.05 -> 87.09Hurt
use DID instead of RADD at few denoising steps (16)RADD small, OWT fixed length · generative PPL, 16 stepsgenerative PPL: 284.78 -> 158.93Improved
use DID instead of RADD at many denoising steps (512)RADD small, OWT fixed length · generative PPL, 512 stepsgenerative PPL: 84.00 -> 86.04Hurt
use DID instead of RADD in variable-length settingRADD small, Stories · generative PPL, 64 stepsgenerative PPL: 81.92 -> 22.78Improved
use DID instead of RADD (removing MASK tokens) for inferenceRADD small, OWT · inference time, 512 stepsspeedup: 1.58xImproved
use DID instead of RADD in variable-length setting for trainingRADD large, Stories · training timespeedup: 2.58x -> 3.42x with model sizeImproved
train DID with DISE objective instead of DICE (fixed-length simplification) (ablation)DID · zero-shot language modelingperplexityHurt
train RADD at padding length 512 instead of 1024 (ablation)RADD small, Stories · variable-length generationspeed and quality: ~1.59x slower than DIDHurt

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