Research map / Architectures and efficiency / Attention and transformers
Attention Sparsity is Input-Stable: Training-Free Sparse Attention for Video Generation via Offline Sparsity Profiling and Online QK Co-Clustering
Jiayi Luo, Jiayu Chen, Jiankun Wang, Cong Wang, Hanxin Zhu, Qingyun Sun, Chen Gao, Zhibo Chen, Jianxin Li
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add bidirectional Q-K co-clustering instead of independent K-means partitioning (ablation) | SVOO · Wan2.1-1.3B, Wan2.2-14B, HunyuanVideo T2V | quality | Improved |
| add offline layer-wise sparsity profiling instead of fixed recall threshold 90% (ablation) | SVOO · Wan2.1-1.3B, Wan2.2-14B, HunyuanVideo T2V | efficiency | Improved |
| use SVOO instead of SVG2 | sparse attention · Wan2.1-1.3B-T2V | speedup: 1.93x vs 1.73x; outperforming the runner-up SVG2 by 0.20x | Improved |
| use SVOO instead of SVG2 | sparse attention · Wan2.1-1.3B-T2V | image quality: 4.74% | Improved |
| use SVOO instead of SVG2 | sparse attention · Wan2.1-1.3B-T2V | aesthetic quality: 4.30% | Improved |
| increase sparsity level | SVOO · random VBench subset | generation quality | Hurt |
| reuse clustering results, recompute every 20 steps | SVOO · text-to-video | quality | No clear effect |
| use bidirectional co-clustering instead of K-means (ablation) | attention recall · attention recall vs SVG2 | attention recall | Improved |
| SVOO vs SVG2 PSNR on HunyuanVideo T2V | SVOO · HunyuanVideo-T2V | PSNR: 24.879 vs 25.218 | Hurt |
Closest papers by meaning
- Sparse Video-Gen: Accelerating Video Diffusion Transformers with Spatial-Temporal Sparsity (ICML 2025 · similarity 0.92)
- Bidirectional Sparse Attention for Faster Video Diffusion Training (NeurIPS 2026 · similarity 0.92)
- iCATS: Fast Video Generation via Interaction-Aware Sparse Attention and Timestep-Adaptive Sparsity (NeurIPS 2026 · similarity 0.91)
- Faster Video Diffusion with Trainable Sparse Attention (NeurIPS 2025 · similarity 0.91)
- HyperVAttention: Efficient Sparse Attention with Spatio-Temporal Clustering for Video Diffusion (NeurIPS 2026 · similarity 0.91)
- PISA: Piecewise Sparse Attention Is Wiser for Efficient Diffusion Transformers (NeurIPS 2026 · similarity 0.90)
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.
