תייחוס רציף עבור ייצור וידאו יעיל

תקציר מקורי באנגליתarXiv:2609.37001v1 Announce Type: cross Abstract: Diffusion Transformers (DiTs) enable high-quality video generation but suffer from substantial inference latency, primarily attributable to the computationally expensive full spatio-temporal attention. While sparse attention methods offer potential solutions, existing approaches face an inherent flexibility--efficiency dilemma: predefined masks lack the flexibility to capture diverse attention patterns, while runtime-determined masks introduce overheads and sacrifice hardware efficiency. We identify the lack of a unified structural characterization of DiT attention as a key limitation of existing methods, and establish that video DiT attention exhibits \textbf{periodic diagonal stripe structures} along both temporal and spatial dimensions.