fixed the magi-1 link

Author: Strivin0311

_pages/dffa.md

@@ -93,7 +93,7 @@ _styles: >
 
 Training large-scale models for video generation presents two major challenges: (1) The extremely long context length of video tokens, which reaching up to 4 million during training, results in prohibitive computational and memory overhead. (2) The combination of block-causal attention and Packing-and-Padding (PnP) introduces highly complex attention mask patterns. 
 
-To address these challenges, we propose [MagiAttention](https://github.com/SandAI-org/MagiAttention), which aims to support a wide variety of attention mask types with **kernel-level flexibility**, while achieving **linear scalability** with respect to context-parallel (CP) size across a broad range of scenarios, particularly suitable for training tasks involving <u><em>ultra-long, heterogeneous mask</em></u> training like video-generation for [Magi-1](https://github.com/SandAI-org/Magi-1).
+To address these challenges, we propose [MagiAttention](https://github.com/SandAI-org/MagiAttention), which aims to support a wide variety of attention mask types with **kernel-level flexibility**, while achieving **linear scalability** with respect to context-parallel (CP) size across a broad range of scenarios, particularly suitable for training tasks involving <u><em>ultra-long, heterogeneous mask</em></u> training like video-generation for [Magi-1](https://github.com/SandAI-org/MAGI-1).
 
 
 ## Introduction