VideoNSA: Native Sparse Attention Scales Video Understanding
Enxin Song, Wenhao Chai, Shusheng Yang, Ethan J. Armand, Xiaojun Shan, Haiyang Xu, Jianwen Xie, Zhuowen Tu
Abstract
Video understanding in multimodal language models remains limited by context length: models often miss key transition frames and struggle to maintain coherence across long time scales. To address this, we adapt Native Sparse Attention (NSA) to video-language models. **Our method, VideoNSA, adapts Qwen2.5-VL through end-to-end training on a 216K video instruction dataset. We employ a hardware-aware hybrid approach to attention, preserving dense attention for text, while employing NSA for video.** Compared to token-compression and training-free sparse baselines, VideoNSA achieves improved performance on long-video understanding, temporal reasoning, and spatial benchmarks. Further ablation analysis reveals four key findings: (1) reliable scaling to 128K tokens; (2) an optimal global–local attention allocation at a fixed budget; (3) task-dependent branch usage patterns; and (4) the learnable combined sparse attention help induce dynamic attention sinks.
BibTeX
@inproceedings{
song2026videonsa,
title={Video{NSA}: Native Sparse Attention Scales Video Understanding},
author={Enxin Song and Wenhao Chai and Shusheng Yang and Ethan J. Armand and Xiaojun Shan and Haiyang Xu and Jianwen Xie and Zhuowen Tu},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=zA2LbsUMDd}
}