English

S$^2$Q-VDiT: Accurate Quantized Video Diffusion Transformer with Salient Data and Sparse Token Distillation

Computer Vision and Pattern Recognition 2026-03-10 v4

Abstract

Diffusion transformers have emerged as the mainstream paradigm for video generation models. However, the use of up to billions of parameters incurs significant computational costs. Quantization offers a promising solution by reducing memory usage and accelerating inference. Nonetheless, we observe that the joint modeling of spatial and temporal information in video diffusion models (V-DMs) leads to extremely long token sequences, which introduces high calibration variance and learning challenges. To address these issues, we propose S2^2Q-VDiT, a post-training quantization framework for V-DMs that leverages Salient data and Sparse token distillation. During the calibration phase, we identify that quantization performance is highly sensitive to the choice of calibration data. To mitigate this, we introduce \textit{Hessian-aware Salient Data Selection}, which constructs high-quality calibration datasets by considering both diffusion and quantization characteristics unique to V-DMs. To tackle the learning challenges, we further analyze the sparse attention patterns inherent in V-DMs. Based on this observation, we propose \textit{Attention-guided Sparse Token Distillation}, which exploits token-wise attention distributions to emphasize tokens that are more influential to the model's output. Under W4A6 quantization, S2^2Q-VDiT achieves lossless performance while delivering 3.9×3.9\times model compression and 1.3×1.3\times inference acceleration. Code will be available at https://github.com/wlfeng0509/s2q-vdit.

Keywords

Cite

@article{arxiv.2508.04016,
  title  = {S$^2$Q-VDiT: Accurate Quantized Video Diffusion Transformer with Salient Data and Sparse Token Distillation},
  author = {Weilun Feng and Haotong Qin and Chuanguang Yang and Xiangqi Li and Han Yang and Yuqi Li and Zhulin An and Libo Huang and Michele Magno and Yongjun Xu},
  journal= {arXiv preprint arXiv:2508.04016},
  year   = {2026}
}

Comments

Accepted by NeurIPS 2025

R2 v1 2026-07-01T04:36:25.581Z