npu-smi info：

apiVersion: mindxdl.gitee.com/v1
kind: AscendJob
metadata:
  name: ascend-training-job
  namespace: xingsuan
  labels:
    app: ascend-training-job
    framework: pytorch
spec:
  replicaSpecs:
    Master:
      replicas: 1
      template:
        metadata:
          labels:
            app: ascend-master
            framework: pytorch
        spec:
          nodeSelector:
            host-arch: huawei-arm
            accelerator-type: module-310b-8
          containers:
            - name: ascend
              image: ascend-pytorch:23.0.0-A1-1.11.0-ubuntu18.04
              securityContext:
                runAsUser: 0
                runAsGroup: 0
              command:
                - /bin/bash
                - "-c"
              args: [ "cd /job/code/scripts; bash train_start.sh /job/code /job/output main.py --data=/job/data/resnet50/imagenet --amp --arch=resnet50 --seed=49 -j=128 --world-size=1 --lr=1.6 --dist-backend='hccl' --multiprocessing-distributed --epochs=90 --batch-size=4096"]
              resources:
                limits:
                  huawei.com/Ascend310: 1
                  cpu: "2"
                  memory: "4Gi"
                requests:
                  huawei.com/Ascend310: 1
                  cpu: "2"
                  memory: "4Gi"