https://wavespeed.ai/ Context parallel attention that accelerates DiT model inference with dynamic caching
