[AMD] Add DLLM support for AMD GPUs with LLaDA2 testing

[sunxxuns]

enable basic text diffusion model for amd cards.

Launch Command

python -m sglang.launch_server \
--model-path inclusionAI/LLaDA2.0-mini \
--dllm-algorithm LowConfidence \
--trust-remote-code \
--mem-fraction-static 0.9 \
--max-running-requests 1 \
--attention-backend triton \
--host 0.0.0.0 \
--port 30000

Test Command

curl -s http://127.0.0.1:30000/v1/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "inclusionAI/LLaDA2.0-mini",
    "prompt": "Write a 1-sentence summary of what SGLang is.",
    "temperature": 0.2,
    "max_tokens": 64,
    "stream": false
  }' | jq .

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[ClawSeven]

@sunxxuns Hi, great work on adding dLLM support for AMD! Your code can serve as a reference for other hardware implementations. I’ve added the link to the dLLM roadmap so others can follow your approach.

[yushengsu-thu]

The main added parts are in CI/D and will not affect the functions on other hardware. but might need to rebase

[sunxxuns]

The main added parts in on CI/D and will not affect the functions on other hardware. but might need to rebase

rebased, pls merge if no issue to avoid next rebase.

[zyzshishui]

/rerun-failed-ci