AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Self-Speculative Decoding Accelerates Lossless Inference in Any-Order and Any-Subset Autoregressive Models

Gabe Guo, Stefano Ermon

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding

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
use any-subset speculative decoding with self as draftfinetuned AS-ARM · WikiText 95% maskedModel NFE: 486.0 -> 434.1Improved
use any-subset speculative decoding with self as draftfinetuned AS-ARM · WikiText 95% maskedGen PPL: 107.9 -> 107.6No clear effect
use context n-gram draft instead of self draftASSD · WikiText 95% maskedTime (s): 16.50 -> 16.80Hurt
use speculative decoding on off-the-shelf XLNetXLNet · WikiText 95% maskedTime (s): 18.04 -> 9.36Improved
speculate only k=2 tokens (ablation)ASSD · WikiText length 256speedup (%): 8.8 -> -2.7Hurt
increase speculated tokens k from 4 to 20 (ablation)ASSD · WikiText length 128speedup (%): 8.8 -> 9.0No clear effect
increase sequence length from 128 to 1024 (ablation)ASSD · WikiText 95% maskedspeedup (%): 8.94 -> 11.00Improved
add KV cache (ablation)XLNet 340M ASSD · infill 128 of 1024time per NFE (s): 0.1407 -> 0.0706Improved
add epsilon acceptance tolerance 0.1 (ablation)ASSD · WikiText length 256Spec PPL: 122.70 -> 185.26Hurt
add epsilon acceptance tolerance 0.1 (ablation)ASSD · WikiText length 256speedup (%): 8.28 -> 14.22Improved
apply speculative decoding to left-to-right promptingXLNet · WikiText prefix promptingTime (s): 14.3 -> 13.3Improved
finetune on wider masking ratios instead of pretrained weights (ablation)AS-ARM · ROCStories infill 1/5ROUGE-1: 14.4 -> 13.1Hurt

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