Tag
This blog from NVIDIA Research discusses how sequence parallelism can scale long-video training systems for both understanding and generation, addressing the challenge of fitting very long video sequences across multiple GPUs.