Add SGLang usage section to the Cosmos3-Edge model card

#53

Adds an SGLang Diffusion usage section to the model card.

Contents:

  • Install block and serve command. Cosmos3-Edge fits comfortably on one GPU, so --dit-cpu-offload false keeps the transformer resident; additional GPUs use Ulysses sequence parallelism or CFG parallelism.
  • Table of the supported endpoints (text-to-image, text-to-video, image-to-video).
  • The per-mode defaults SGLang applies (640x640 / guidance 7.0 for text-to-image, 832x480 / guidance 5.0 for video, flow shift 3.0) and the list of supported sizes, since the model is trained at 256p and 480p.
  • Image-to-video, text-to-video and text-to-image examples, verified end-to-end.

Please do not merge yet - holding until Cosmos 3 support is merged into SGLang: sgl-project/sglang#31590.

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment