[New Model]: Qwen3 support

[Yikun]

The model to consider.

• Qwen/Qwen3-8B
• Qwen/Qwen3-MoE-15B-A2B

The closest model vllm already supports.

No response

What's your difficulty of supporting the model you want?

First priority:

• Add CI for Qwen/Qwen3-0.6B: [CI] Add Qwen3-0.6B-Base test #717
• Download models and run test (functional / accuray / perf) for Qwen/Qwen3-8B
  • functional: @wangxiyuan
    • offline test
    • online test
  • accruracy: @hfadzxy
  • perf: @@Potabk
• Update Single NPU (Qwen2.5 7B) to Single NPU (Qwen3 8B): Update installation and tutorial doc #711
• Announcement on wechat post on Open Source Now: 使用vLLM Ascend 部署 Qwen3
• Validate on Qwen3 docker image:
  • 0.8.4rc2
  • 0.8.4rc2-openeuler

Second priority:

• Fix MOE error: [Model] Support common fused moe ops for moe model #709
• Add CI for Qwen/Qwen3-MoE-15B-A2B
• Download models and run test (functional / accuray / perf) for Qwen/Qwen3-MoE-15B-A2B
• Announcement on wechat post on Open Source Now: 使用 vLLM Ascend 部署 Qwen/Qwen3-MoE

[wangxiyuan]

qwen3 works now