AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

SparseD: Sparse Attention for Diffusion Language Models

Zeqing Wang, Gongfan Fang, Xinyin Ma, Xingyi Yang, Xinchao Wang

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 14 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 skipping sparse attention in early denoising steps (ablation)SparseD (LLaDA-1.5) · RULER-4kaccuracy: 87.91 -> 90.89Improved
add skipping sparse attention in early denoising steps (ablation)SparseD (LLaDA-1.5) · RULER 64k sample latencylatency (s): 1552 -> 1695Hurt
add sparse pattern reusing across denoising steps (ablation)SparseD (LLaDA-1.5) · RULER 64k sample latencylatency (s): 30020 -> 1695Improved
add sparse pattern reusing across denoising steps (ablation)SparseD (LLaDA-1.5) · RULER-4kaccuracy: 90.82 -> 90.89No clear effect
add isolated selection of prefill and generation tokens (ablation)SparseD (LLaDA-1.5) · RULER-4kaccuracy: 90.53 -> 90.89Improved
add isolated selection of prefill and generation tokens (ablation)SparseD (LLaDA-1.5) · RULER 64k sample latencylatency (s): 1687 -> 1695No clear effect
use SparseD instead of StreamingLLMDream-7B-Instruct · Average of MMLU, GSM8k, HumanEval, RULER-4k, RULER-8kaccuracy (avg): 50.17 -> 72.38Improved
use SparseD instead of dKV-Cache on long contextDream-7B-Instruct · RULER-8kaccuracy: 55.08 -> 72.47Improved
use SparseD instead of Fast-dLLM on long contextLLaDA-1.5 · RULER-8kaccuracy: 47.76 -> 62.44Improved
apply SparseD to LLaDA-1.5LLaDA-1.5 · Average of MMLU, GSM8k, HumanEval, RULER-4k, RULER-8kaccuracy (avg): 67.33 -> 67.62No clear effect
apply SparseD to Dream-7B-InstructDream-7B-Instruct · Average of MMLU, GSM8k, HumanEval, RULER-4k, RULER-8kaccuracy (avg): 72.42 -> 72.38No clear effect
increase diffusion steps from 128 to 1024SparseD (Dream-7B-Instruct) · 64k contextspeedup vs FlashAttention: 1.23x -> 1.50xImproved

Later papers that compared with this method

2 later accepted papers report SparseD as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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