Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.16515 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912997314658304 |
|---|---|
| author | Fang, Tongcheng Zhang, Hanling Xie, Ruiqi Han, Zhuo Tao, Xin Zhao, Tianchen Wan, Pengfei Ding, Wenbo Ouyang, Wanli Ning, Xuefei Wang, Yu |
| author_facet | Fang, Tongcheng Zhang, Hanling Xie, Ruiqi Han, Zhuo Tao, Xin Zhao, Tianchen Wan, Pengfei Ding, Wenbo Ouyang, Wanli Ning, Xuefei Wang, Yu |
| contents | Diffusion Transformers have demonstrated remarkable performance in video generation. However, their long input sequences incur substantial latency due to the quadratic complexity of full attention. Various sparse attention mechanisms have been proposed. Training-free approaches are limited to moderate sparsity and thus yield only modest acceleration, whereas training-based methods can reach much higher sparsity but demand substantial data and computation. In this work, we propose SALAD, introducing a lightweight linear attention branch in parallel with the sparse attention. Leveraging a Multi-level Static-Dynamic Scaling Strategy to balance the two branches, our method attains up to 90% sparsity and 1.52-2.03x inference speedup across different models and sequence lengths, while maintaining generation quality comparable to the full attention baseline. Moreover, our finetuning process is highly efficient, requiring only 2,000 video samples, fewer than 1,600 training steps, and no more than 30 GPU hours with a batch size of 8. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_16515 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SALAD: Achieve High-Sparsity Attention via Efficient Linear Attention Tuning for Video Diffusion Transformer Fang, Tongcheng Zhang, Hanling Xie, Ruiqi Han, Zhuo Tao, Xin Zhao, Tianchen Wan, Pengfei Ding, Wenbo Ouyang, Wanli Ning, Xuefei Wang, Yu Computer Vision and Pattern Recognition Diffusion Transformers have demonstrated remarkable performance in video generation. However, their long input sequences incur substantial latency due to the quadratic complexity of full attention. Various sparse attention mechanisms have been proposed. Training-free approaches are limited to moderate sparsity and thus yield only modest acceleration, whereas training-based methods can reach much higher sparsity but demand substantial data and computation. In this work, we propose SALAD, introducing a lightweight linear attention branch in parallel with the sparse attention. Leveraging a Multi-level Static-Dynamic Scaling Strategy to balance the two branches, our method attains up to 90% sparsity and 1.52-2.03x inference speedup across different models and sequence lengths, while maintaining generation quality comparable to the full attention baseline. Moreover, our finetuning process is highly efficient, requiring only 2,000 video samples, fewer than 1,600 training steps, and no more than 30 GPU hours with a batch size of 8. |
| title | SALAD: Achieve High-Sparsity Attention via Efficient Linear Attention Tuning for Video Diffusion Transformer |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.16515 |