MC-Sparse: Deconstructing and Closing the Dense-Sparse Attention Gap in Diffusion Transformers
[Widzenie komputerowe] Autorzy: Jiarui Chen, Zeqiang Lai, Jiangshan Wang i 5 więcej
Sparse attention is a primary approach to reducing the latency of diffusion transformers in long-sequence generation tasks, such as video and high-resolution 3D asset generation. However, existing methods can degrade generation quality and fidelity at high sparsity levels. Through controlled oracle comparisons, we trace this degradation to three sources: constraints imposed by token grouping, inaccurate interaction selection, and the attention contributions lost when tokens are discarded. Guided by this analysis, we propose Meta-Cached Sparse Attention (MC-Sparse), a training-free framework that selects individual key-value (KV) tokens while organizing similar queries into tile-aligned groups for efficient GPU execution. MC-Sparse caches metadata comprising query groups, KV indices selected using exact attention probabilities, and residuals between dense and sparse attention outputs, an…
- Kategoria arxiv
- cs.CV
- Kategoria
- Widzenie komputerowe