Create Mxnet Task

Last Updated：2022-01-14

You can create a Mxnet task.

Prerequisites

You already install the CCE AI Job Scheduler and CCE Deep Learning Frameworks Operator components successfully. Otherwise, the cloud native AI feature is unavailable.
If you are a sub-user, you can only use the queue to create a task if you are among the users associated with the queue.
During the installation of the component CCE Deep Learning Frameworks Operator, the system is installed with the Mxnet deep learning framework.

Operation Steps

Log in to Baidu AI Cloud Official Website, and then enter the management console.
Select “Product Service > Cloud Native > CCE”, and click CCE to enter the container engine management console.
Click Cluster Management > Cluster List in the navbar on the left side.
On the cluster list page, click the target cluster name to enter the cluster management page.
On the cluster management page, click Cloud Native AI > Task Management.
On the task management page, click Create Task.
On the basic information, complete the configuration of the task.

Task name: Customize the task name, which supports uppercase and lowercase letters, numbers, -, _, /, ., and other special characters, must start with a Chinese character or letter and have a length of 1-65 characters.
Queue: Select the queue associated with the new task.
Framework: Select the deep learning framework "Mxnet" corresponding to the task.

Complete the configuration by referring to the yaml template:

apiVersion: "kubeflow.org/v1"
kind: "MXJob"
metadata:
  name: "mxnet-job"
spec:
  jobMode: MXTrain
  mxReplicaSpecs:
    Scheduler:
      replicas: 1
      restartPolicy: Never
      template:
        metadata:
          annotations:
            sidecar.istio.io/inject: "false"
            # if your libcuda.so.1 is in custom path, set the correct path with the following annotation
            # kubernetes.io/baidu-cgpu.nvidia-driver-lib: /usr/lib64
        spec:
          schedulerName: volcano
          containers:
            - name: mxnet
              image: registry.baidubce.com/cce-public/mxjob/mxnet:gpu
              resources:
                limits:
                  baidu.com/v100_32g_cgpu: "1"
                  # for gpu core/memory isolation
                  baidu.com/v100_32g_cgpu_core: 5
                  baidu.com/v100_32g_cgpu_memory: "1"
              # if gpu core isolation is enabled, set the following preStop hook for graceful shutdown.
              # `train_mnist.py` needs to be replaced with the name of your gpu process.
              lifecycle:
                preStop:
                  exec:
                    command: ["/bin/sh", "-c", "kill -10 `ps -ef | grep train_mnist.py | grep -v grep | awk '{print $2}'` && sleep 1"]
    Server:
      replicas: 1
      restartPolicy: Never
      template:
        metadata:
          annotations:
            sidecar.istio.io/inject: "false"
            # if your libcuda.so.1 is in custom path, set the correct path with the following annotation
            # kubernetes.io/baidu-cgpu.nvidia-driver-lib: /usr/lib64
        spec:
          schedulerName: volcano
          containers:
            - name: mxnet
              image: registry.baidubce.com/cce-public/mxjob/mxnet:gpu
              resources:
                limits:
                  baidu.com/v100_32g_cgpu: "1"
                  # for gpu core/memory isolation
                  baidu.com/v100_32g_cgpu_core: 5
                  baidu.com/v100_32g_cgpu_memory: "1"
              # if gpu core isolation is enabled, set the following preStop hook for graceful shutdown.
              # `train_mnist.py` needs to be replaced with the name of your gpu process.
              lifecycle:
                preStop:
                  exec:
                    command: ["/bin/sh", "-c", "kill -10 `ps -ef | grep train_mnist.py | grep -v grep | awk '{print $2}'` && sleep 1"]
    Worker:
      replicas: 1
      restartPolicy: Never
      template:
        metadata:
          annotations:
            sidecar.istio.io/inject: "false"
            # if your libcuda.so.1 is in custom path, set the correct path with the following annotation
            # kubernetes.io/baidu-cgpu.nvidia-driver-lib: /usr/lib64
        spec:
          schedulerName: volcano
          containers:
          - name: mxnet
            image: registry.baidubce.com/cce-public/mxjob/mxnet:gpu
            command: ["python"]
            args: ["/incubator-mxnet/example/image-classification/train_mnist.py","--num-epochs","10","--num-layers","2","--kv-store","dist_device_sync","--gpus","0"]
            resources:
              requests:
                cpu: 1
                memory: 1Gi
              limits:
                baidu.com/v100_32g_cgpu: "1"
                # for gpu core/memory isolation
                baidu.com/v100_32g_cgpu_core: 20
                baidu.com/v100_32g_cgpu_memory: "4"
            # if gpu core isolation is enabled, set the following preStop hook for graceful shutdown.
            # `train_mnist.py` needs to be replaced with the name of your gpu process.
            lifecycle:
              preStop:
                exec:
                  command: ["/bin/sh", "-c", "kill -10 `ps -ef | grep train_mnist.py | grep -v grep | awk '{print $2}'` && sleep 1"]

Click the “OK” button to complete the task creation.

Queue Management

Create PaddlePaddle Task

百度智能云

Cloud Container Engine

Create Mxnet Task

Prerequisites

Operation Steps