Search papers, labs, and topics across Lattice.
OSP-Next, a text-to-video generation model, tackles the quadratic cost of full attention in Diffusion Transformers by using a hybrid full-sparse attention architecture with Skiparse-2D Attention. They introduce Sparse Sequence Parallelism (SSP) to efficiently parallelize sparse attention, reducing communication volume by 75% compared to Ulysses Sequence Parallelism. OSP-Next further incorporates HiF8 quantization and Mix-GRPO post-training, achieving a VBench total score of 83.73% and significant speedups on both NVIDIA H200 and Ascend 950PR GPUs.
Achieve near state-of-the-art video generation quality with OSP-Next, while realizing up to 2.27x speedups on Ascend 950PR GPUs through a novel combination of sparse attention, quantization, and parallelism techniques.
Diffusion Transformers achieve strong video generation quality, but the quadratic cost of full attention limits efficiency. We introduce OSP-Next, an efficient text-to-video generation model that integrates sparse attention, parallelism, quantization, and reinforcement learning. OSP-Next uses a hybrid full-sparse attention architecture, where the sparse component is implemented with Skiparse-2D Attention. This fixed-pattern mechanism applies token-wise and group-wise sparse attention along spatial dimensions, leveraging locality while maintaining native compatibility with FlashAttention kernels. Based on the local equivalence of rearrangement in Skiparse-2D Attention, we further propose Sparse Sequence Parallelism (SSP), which partitions subsequences across ranks and switches sparse patterns through a single All-to-All communication. Compared with Ulysses Sequence Parallelism (SP), SSP provides a native parallel strategy for sparse attention and reduces communication volume by 75%. OSP-Next also incorporates HiF8 quantization to enable stable joint training with 8-bit quantization and sparse fine-tuning, and applies Mix-GRPO post-training to improve the performance of the sparse model. Experiments show that OSP-Next achieves a VBench total score of 83.73%, surpassing the Wan2.1 baseline. Under the 5-second 720P and 5-second 768P settings, OSP-Next achieves up to 1.64times single-GPU speedup and over 1.52times eight-GPU speedup on NVIDIA H200 GPUs. In addition, with only a 0.4% drop in VBench total score, OSP-Next-HiF8 achieves 1.69times and 2.27times speedups under the two settings on a single Ascend 950PR, demonstrating the efficiency and performance of OSP-Next across hardware platforms.
