[Feature] Support DeepSeek Janus Models

[juliancoy]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

Docker is a valuable tool for the management of dependencies. Indeed, it can simplify the running of Janus Models to a single command:

docker run -it --rm \
  -p 8000:8000 \
  -d \
  -v huggingface:/root/.cache/huggingface \
  -w /app \
  --gpus all \
  --name janus \
  -e MODEL_NAME=deepseek-ai/Janus-Pro-7B \
  julianfl0w/janus:latest

Make sure it's working by navigating in your browser to
http://localhost:8000/webui

and by running

docker logs janus

This keeps all the Torch dependencies contained within the image, meaning the user doesn't have to adjust their base installations to run models like these.

Note: You will have to install NVIDIA Container Runtime (or equivalent)

The implementation of this Dockerfile can be found at DeepSeek Janus PR#38

Related resources

No response

[zhaochenyang20]

@mickqian Thanks for your interests