Ascend 910B 快速部署Qwen3-VL-30B-A3B-Instruct环境8张32GB显存大910B下载模型权重pipinstallmodelscope modelscope download--modelQwen/Qwen3-VL-30B-A3B-Instruct安装运行Docker容器# Update the vllm-ascend imageexportIMAGEquay.io/ascend/vllm-ascend:v0.18.0dockerrun--rm\--namevllm-ascend\--shm-size1g\--nethost\--device/dev/davinci0\--device/dev/davinci1\--device/dev/davinci_manager\--device/dev/devmm_svm\--device/dev/hisi_hdc\-v/usr/local/dcmi:/usr/local/dcmi\-v/usr/local/bin/npu-smi:/usr/local/bin/npu-smi\-v/usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/\-v/usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info\-v/etc/ascend_install.info:/etc/ascend_install.info\-v/root/.cache:/root/.cache\-v/data:/data\-vpath/to/your/media:/media\-it$IMAGEbash设置环境变量# Load model from ModelScope to speed up downloadexportVLLM_USE_MODELSCOPETrue# Set max_split_size_mb to reduce memory fragmentation and avoid out of memoryexportPYTORCH_NPU_ALLOC_CONFmax_split_size_mb:256启动服务vllm serve Qwen/Qwen3-VL-30B-A3B-Instruct\--tensor-parallel-size2\--enable-expert-parallel\--limit-mm-per-prompt.video0\--max-model-len128000通过接口访问服务curlhttp://localhost:8000/v1/chat/completions\-HContent-Type: application/json\-d{ model: Qwen/Qwen3-VL-30B-A3B-Instruct, messages: [ {role: system, content: 你是一个图片识别助手.}, {role: user, content: [ {type: image_url, image_url: base64图片}, {type: text, text: 请帮我分析这张图片中是什么?} ]} ], max_completion_tokens: 100 }可以通过这个网站来查看更详细的使用方法。