LASP-2: Rethinking Sequence Parallelism for Linear Attention and Its Hybrid

Open in new window