RoPeSLR: Sparse-Low-Rank Attention for Diffusion Transformers

Created on May 27, 2026

2026   ·   diffusion-transformers   video-generation   efficient-training   ·   research

RoPeSLR is a 3D RoPE-driven sparse-low-rank attention framework for efficient Diffusion Transformers. It decouples attention into sparse and low-rank components while retaining positional structure for visual generation.



Enjoy Reading This Article?

Here are some more articles you might like to read next:

  • SparkDiffusion: Mitigating the High-Sparsity Trap
  • CrossDistill: Balancing Quality and Diversity
  • RoLA: Rotary-Positioned Low-Rank Linear Attention
  • CR-Net: Cross-Layer Low-Rank Structure
  • Mixture of Distributions Matters