perf(wan2.2): Optimize attention kernels, rematerialization, and Ulysses parallelism for WAN 2.2 Training on TPU 7x - #487

perf(wan2.2): Optimize attention kernels, rematerialization, and Ulysses parallelism for WAN 2.2 Training on TPU 7x#487

copybara-service[bot] merged 1 commit into

AI-Hypercomputer:mainAI-Hypercomputer/maxdiffusion:mainfrom

Toshi-31:wan-2.2-perf-optimizationsToshi-31/maxdiffusion:wan-2.2-perf-optimizationsCopy head branch name to clipboard