cs.CVNov 14, 2026

dit.xpu : Taming DiT Sparse Inference on Mobile SoCs

Authors: Jiayu Chen, Jinye Hu, Maoliang Li, Anqian Cen, Hailong Zou, Zihao Zheng, Xinming Wei, Guojie Luo, 

Organizations: Peking University, School of Computer Science, Beijing, China · Peking University, School of Electronics Engineering and Computer Science, Beijing, China · Beijing Institute of Technology, School of Computer Science & Technology, Beijing, China · Peking University, Beijing, China

Abstract

Diffusion Transformers (DiTs) enable on-device image generation and editing but incur high attention costs at high resolutions. Existing sparse attention reduces computation, yet dynamic sparsity introduces preprocessing and irregular data accesses that poorly fit mobile NPUs. We present DiT.XPU, a sparse DiT inference framework for heterogeneous mobile SoCs. DiT.XPU combines tile-aligned sparse attention with anchor-based error compensation by exploiting slowly varying dense-sparse output residuals across denoising steps. Heterogeneous XPU collaboration overlaps NPU dense attention with GPU sparse calibration, while mask-guided dataflow reduces data transfers and layout conversion by overlapping data movement with computation. Across five DiT models on two Snapdragon smartphones, DiT.XPU achieves up to 1.56x end-to-end denoising speedup over the Hexagon-NPU backend and 11.5x attention-kernel speedup over the fastest alternative, while preserving generation quality close to dense attention. Code is available at https://github.com/IF-LAB-PKU/dit.xpu.

Figures & tables

Explore similar work

CardsList
  1. EdgeDiT: Hardware-Aware Diffusion Transformers for Efficient On-Device Image Generation

    Mar 30, 2026Sravanth Kodavanti, Manjunath Arveti, Sowmya Vajrala +2Diffusion TransformerDiffusion-Based Image Generation

  2. SnapGen++: Unleashing Diffusion Transformers for Efficient High-Fidelity Image Generation on Edge Devices

    Jan 13, 2026Dongting Hu, Aarush Gupta, Magzhan Gabidolla +12Diffusion TransformerDiffusion Model Distillation

  3. Sink-Sparse Hybrid Attention for Efficient Inference in T2I/T2V Diffusion Transformers

    Oct 16, 2026Bingxin Xing, Yu Zhang, Jiatong Ding +2Diffusion Model Inference AccelerationSparse Attention