华为昇腾MindIE部署Qwen2.5-72B模型启动卡住，日志中断在HCCL初始化阶段求助

[2026-01-29 10:15:23.456+08:00] [12345] [12346] [server] [INFO] [share_memory.cpp:168] : [ShareMemory::SharedMemorySizeCheck] shared memory size check success.
[2026-01-29 10:15:23.568+08:00] [12347] [12347] [server] [DEBUG] [main.cpp:286] : [Connector] process 12347 belong to process group: 12345
[2026-01-29 10:15:23.569+08:00] [12347] [12347] [server] [DEBUG] [slave_IPC_communicator.cpp:747] : [SetupMsgChannels]slave communicator lanuch threads success.
[2026-01-29 10:15:24.336+08:00] [12347] [12351] [server] [DEBUG] [slave_IPC_communicator.cpp:378] : [=========== Backend rank 0, start processing the broadcast message =============]
[2026-01-29 10:15:24.337+08:00] [12347] [12351] [server] [DEBUG] [slave_IPC_communicator.cpp:399] : [=========== Backend rank 0 is Single Node Infer =============]

# 安装CANN 2.1.RC2
sudo yum install ascend-devel-2.1.rc2 -y

# 拉取MindIE官方镜像
docker pull swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:2.1.RC3-800I-A2-py311-openeuler24.03-lts

docker run -it -d --net=host --shm-size=4g \
    --name qwen25-72b \
    --device=/dev/davinci_manager \
    --device=/dev/hisi_hdc \
    --device=/dev/devmm_svm \
    --device=/dev/davinci0 --device=/dev/davinci1 \
    --device=/dev/davinci2 --device=/dev/davinci3 \
    -e ASCEND_VISIBLE_DEVICES=0,1,2,3 \
    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver:ro \
    -v /data/models/Qwen2.5-72B-Instruct:/models:rw \
    swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:2.1.RC3-800I-A2-py311-openeuler24.03-lts

{
  "ServerConfig": {
    "ipAddress": "0.0.0.0",
    "port": 7025,
    "managementPort": 7026,
    "allowAllZeroIpListening": true,
    "httpsEnabled": false,
    "interCommTLSEnabled": false
  },
  "BackendConfig": {
    "npuDeviceIds": [[0,1,2,3]],
    "modelInstanceNumber": 1,
    "ModelDeployConfig": {
      "maxSeqLen": 8192,
      "maxInputTokenLen": 6144,
      "ModelConfig": [{
        "modelInstanceType": "Standard",
        "modelName": "qwen2.5",
        "modelWeightPath": "/models",
        "worldSize": 4,
        "cpuMemSize": 16,
        "npuMemSize": -1,
        "backendType": "atb",
        "trustRemoteCode": true,
        "useCustomKernels": true
      }]
    }
  }
}