AI Paper Atlas

Research map / Video / Video generation

BLADE: Block-Sparse Attention Meets Step Distillation for Efficient Video Generation

Youping Gu, XIAOLONG LI, Yuhao Hu, Chen Minqi, Bohan Zhuang

ICLR 2026 · Video › Video generation · approach: Efficient video diffusion models · 12 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 global tokens (mean-pooled K/V) to ASA (ablation)ASA on CogVideoX-5B · VBench-2.0Total score: 0.539 -> 0.569Improved
add additive ln(n) mask for global tokens (ablation)ASA_G on CogVideoX-5B · VBench-2.0Total score: 0.559 -> 0.569Improved
add Gilbert space-filling-curve token rearrangement (ablation)ASA distilled CogVideoX-5B · VBench-1.0quality score: 0.779 -> 0.788Improved
replace dense FA2 with ASA in 8-step modelWan2.1-1.3B · End-to-end timeseconds: 36.11 -> 24.00Improved
replace dense attention with ASA kernelWan2.1-1.3B · Kernel timems: 73.25 -> 22.21Improved
use ASA instead of STA, training-free, sparsity ~0.75Wan2.1-1.3B 8-step · PSNRPSNR: 16.72 -> 19.55Improved
use ASA instead of SVG, training-free, sparsity ~0.75Wan2.1-1.3B 8-step · SSIMSSIM: 0.6390 -> 0.7433Improved
use ASA instead of Radial Attention at sparsity 0.50Wan2.1-1.3B 8-step · PSNRPSNR: 22.07 -> 22.20Improved
sparsity-aware distillation (ASA_G, 8 steps) vs 50-step dense baselineCogVideoX-5B · VBench-2.0Total score: 0.534 -> 0.569Improved
sparsity-aware distillation (ASA_G, 8 steps) vs 50-step dense baselineWan2.1-1.3B · VBench-2.0Total score: 0.563 -> 0.570No clear effect
use ASA_G instead of dense FA2 in 8-step modelWan2.1-1.3B · VBench-2.0Total score: 0.580 -> 0.570Hurt
use STA sparse attention instead of ASA_GWan2.1-1.3B · VBench-2.0Total score: 0.570 -> 0.528Hurt

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