Diffusion Transformer (DiT), an emerging diffusion model for visual generation, has demonstrated superior perfor mance but suffers from substantial computational costs. Our investigations reveal that these costs primarily stem from the static inference paradig... ...

请注册登录后继续浏览