AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Sparse-dLLM: Accelerating Diffusion LLMs with Dynamic Cache Eviction

Yuerong Song, Xiaoran Liu, Ruixiao Li, Zhigeng Liu, Zengfeng Huang, Qipeng Guo, Ziwei He, Xipeng Qiu

AAAI 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 46 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
apply Sparse-dLLM cache evictionLLaDA-8B-Instruct · GSM8Kthroughput (TPS): 4.57 -> 26.45Improved
apply Sparse-dLLM cache evictionDream-v0-7B-Instruct · GPQAthroughput (TPS): 6.29 -> 32.63Improved
apply Sparse-dLLM cache evictionLLaDA-8B-Instruct · 7-benchmark averageaccuracy (%): 59.24 -> 59.86No clear effect
apply Sparse-dLLM cache evictionDream-v0-7B-Instruct · HumanEvalaccuracy (%): 57.93 -> 50.00Hurt
apply Sparse-dLLM cache evictionDream-v0-7B-Instruct · peak memorypeak memory (GB): 0.9x of vanillaImproved
apply Sparse-dLLM cache evictionLLaDA-8B-Instruct · LongBenchaccuracy (%): 34.55 -> 34.46No clear effect
add bidirectional (prefix and suffix) eviction instead of prefix-only (ablation)Sparse-dLLM on LLaDA-8B-Instruct · GSM8K/MATH/ARC-C averageaccuracy (%): 65.40 -> 66.82Improved
add bidirectional eviction instead of prefix-only (ablation)Sparse-dLLM on Dream-v0-7B-Instruct · GSM8K/MATH/ARC-C averageaccuracy (%): 69.04 -> 69.04No clear effect
increase cache update delay from 1 to 5 steps (ablation)Sparse-dLLM · ARC-Cthroughput (TPS): 36.85 -> 33.78Hurt
increase delay step from 1 to 3 (ablation)Sparse-dLLM · ARC-Caccuracy (%): 88.47 -> 89.49Mixed
raise retention ratio above 0.5 (ablation)Sparse-dLLM · GSM8Kaccuracy (%)No clear effect
change max-pool kernel size away from 3 (ablation)Sparse-dLLM · GSM8Kaccuracy (%)Hurt

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