[Question][Diffusion][SD3.5] Question about SGLang-native support for remaining Diffusers modules for stable-diffusion-3

[Yuxingwang-intel]

Hi SGLang maintainers,

I noticed that the current Stable Diffusion 3.5 code comments mention that the initial implementation directly uses Diffusers' JointTransformerBlock. Could you please share whether there is any near-term plan or timeline to replace the remaining Diffusers-based modules with SGLang-native implementations?

1. transformer.transformer_blocks.*
   • currently uses Diffusers JointTransformerBlock
2. transformer.time_text_embed
   • currently uses Diffusers CombinedTimestepTextProjEmbeddings

Thanks