一、环境依赖

 二、启动容器

示例：

docker run -itd --shm-size=500g --privileged=true --name test0402 \
--privileged=true --net=host \
-v /var/queue_schedule:/var/queue_schedule \
-v /etc/ascend_install.info:/etc/ascend_install.info \
-v /usr/local/sbin:/usr/local/sbin \
-v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
-v /usr/local/Ascend/firmware:/usr/local/Ascend/firmware \
-v /data/models/Qwen/Qwen3___5-122B-A10B:/data \
--device=/dev/davinci_manager:/dev/davinci_manager \
--device=/dev/hisi_hdc:/dev/hisi_hdc \
--entrypoint=bash \
0fa7e4550d22

三、容器内拉起vllm服务

export ASCEND_RT_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
export PYTORCH_NPU_ALLOC_CONF="expandable_segments:True"
export PYTORCH_NPU_ALLOC_CONF=max_split_size_mb:256
export HCCL_IF_IP="10.14.10.222"
export HCCL_OP_EXPANSION_MODE="AIV"
export HCCL_BUFFSIZE=1024
export OMP_NUM_THREADS=1
export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
export TASK_QUEUE_ENABLE=1

export VLLM_ASCEND_ENABLE_PREFETCH_MLP=1
export VLLM_ASCEND_ENABLE_DENSE_OPTIMIZE=1
export VLLM_ASCEND_ENABLE_NZ=1


vllm serve /data \
   --served-model-name "qwen3.5-122b-a10b" \
   --host 10.14.10.222 \
   --port 5678 \
   --tensor-parallel-size 8 \
   --max-model-len 128000 \
   --max-num-batched-tokens 8192 \
   --max-num-seqs 128 \
   --gpu-memory-utilization 0.95 \
   --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY", "cudagraph_capture_sizes":[1,4,8,12,16,24,32,48,56,64]}' \
   --trust-remote-code \
   --enable-auto-tool-choice \
   --tool-call-parser qwen3_coder \
   --async-scheduling \
   --allowed-local-media-path / \
   --mm_processor_cache_type="shm" \
   --mm-processor-cache-gb 0 \
   --speculative-config '{"num_speculative_tokens": 3, "method":"qwen3_5_mtp", "enforce_eager": true}' \
   --additional-config '{"enable_cpu_binding":true, "multistream_overlap_shared_expert": true, "enable_weight_nz_layout":true}'

四、curl请求

curl -H "Accept: application/json" -H "Content-type: application/json" -X POST -d '
{
    "model":"qwen3.5-122b-a10b",
    "messages": [
       {  
          "role": "system", 
          "content": "思考过程和输出结果都用中文"
        },
      {  
          "role": "user", 
          "content": "你是谁"
        }
    ],
   "max_tokens": 32768,
   "presence_penalty":1.5,
   "repetition_penalty":1.0,
   "top_k": 20,
   "top_p": 0.8,
   "temperature": 0.7,
    "chat_template_kwargs": {"enable_thinking": false},
   "stream": true
}' http://10.14.10.222:5678/v1/chat/completions