@misc{indiciae53ae202900e1,
  title = {Decoupling Logical Masks from GPU Execution for Dynamic Block-Sparse Attention},
  author = {Shanghao Liu and Xiaoyun Yu and Wanting Li and Wenqi Jiang},
  year = {2026},
  url = {https://arxiv.org/abs/2609.25869},
  note = {Source identifier: 2609.25869}
}