AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Entropy-Aware Dynamic KV Cache Sparsification for Autoregressive Image Generation and Editing

Tong Tong, Ling Xing, Linjie Li, Rui Yan, Zhengyuan Yang, Lijuan Wang, Alex Jinpeng Wang

ICML 2026 · Image generation › Autoregressive and latent generation · approach: Next-scale and token autoregressive images

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
enable SparseAR entropy-aware KV sparsification at tau=1/6LlamaGen-XL · Text-to-image, FID-30KFID-30K: 5.67 -> 5.55Improved
start sparsification earlier (tau 1/6 -> 1/8) (ablation)SparseAR on Lumina-mGPT-768 · Text-to-image, FID-30KFID-30K: 4.91 -> 10.88Hurt
start sparsification earlier (tau 1/6 -> 1/8) (ablation)SparseAR on Lumina-mGPT-1024 · Text-to-image, FID-30KFID-30K: 3.37 -> 4.08Hurt
start sparsification earlier (tau 1/6 -> 1/8) (ablation)SparseAR on Lumina-mGPT-512 · Text-to-image, FID-30KFID-30K: 6.54 -> 7.22Hurt
use entropy-aware SparseAR instead of Streaming KV compressionBLIP3o-Next image editing · ImgEdit scoreImgEdit: 1.98 -> 3.62Improved
use entropy-aware SparseAR instead of H2OBLIP3o-Next image editing · FID-5kFID-5k: 19.55 -> 13.80Improved
use SparseAR at tau=1/4BLIP3o-Next image editing · Latencylatency: 122.78s -> 111.74sImproved
use fixed pattern compression keeping only condition, initial and recent tokens (ablation)LlamaGen-XL · qualitative text-to-imageimage detailHurt
compress KV in low-entropy regions (entropy 0 to 3) (ablation)Lumina-mGPT-512 · text-to-imagegeneration qualityNo clear effect
compress KV in higher-entropy regions (ablation)Lumina-mGPT-512 · text-to-imagegeneration qualityHurt
lower compression budget to 20% with tau 1/6 or 1/4 (ablation)SparseAR on LlamaGen · FIDFIDMixed

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