AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

ES-dLLM: Efficient Inference for Diffusion Large Language Models by Early-Skipping

Zijian Zhu, Fei Ren, Zhanhong Tan, Kaisheng Ma

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 6 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
early-skipping (ES-dLLM) vs DualCacheES-dLLM · LLaDA-8B GSM8Kthroughput (TPS): 112.15 -> 143.93Improved
ES-dLLM vs vanilla on accuracyES-dLLM · LLaDA-8B MATHaccuracy: 28.14 -> 27.24Hurt
more frequent prompt cache refresh (ES-dLLM*) (ablation)ES-dLLM · LLaDA-8B BBHaccuracy: 54.51 -> 56.66Improved
more frequent prompt cache refresh (ES-dLLM*) (ablation)ES-dLLM · LLaDA-8B BBHthroughput (TPS): 159.89 -> 133.48Hurt
use only confidence (alpha=1) instead of mixing with variation (ablation)ES-dLLM · LLaDA-8B multiple benchmarksaccuracyHurt
use variation only (alpha=0) (ablation)ES-dLLM · LLaDA-8B MATHaccuracyImproved
swap hidden states for key/value as variation indicator (ablation)ES-dLLM · LLaDA-8BaccuracyNo clear effect
skip earlier layer (layer 0 vs layer 16) (ablation)ES-dLLM · LLaDA-8B MATHspeedup vs DualCache: 1.60x vs 1.22x speedup; 26.76 vs 27.52 scoreMixed
larger skip ratio at layer 8 (0.25 to 0.75) (ablation)ES-dLLM · LLaDA-8B MATHspeedup vs DualCache: 1.13x -> 1.70x; score 27.58 -> 27.26Mixed
two skip positions instead of one at layer 4 (equal FLOPs) (ablation)ES-dLLM · LLaDA-8B HumanEvalpass@1: 35.98 -> 37.8Improved
three skip positions instead of two (equal FLOPs) (ablation)ES-dLLM · LLaDA-8B HumanEvalpass@1: 37.8 -> 36.59Hurt
combine with parallel decodingES-dLLM · Dream-7B MBPPthroughput (TPS): 829.71 -> 1469.15Improved

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