4.5. 被配置为使用 RDMA 的 Training Operator PyTorchJob 资源示例

apiVersion: kubeflow.org/v1
kind: PyTorchJob
metadata:
  name: job
spec:
  pytorchReplicaSpecs:
    Master:
      replicas: 1
      restartPolicy: OnFailure
      template:
        metadata:
          annotations:
            k8s.v1.cni.cncf.io/networks: "example-net"
        spec:
          containers:
          - command:
            - /bin/bash
            - -c
            - "your container command"
            env:
            - name: NCCL_SOCKET_IFNAME
              value: "net1"
            - name: NCCL_IB_HCA
              value: "mlx5_1"
            image: quay.io/modh/training:py311-cuda121-torch241
            name: pytorch
            resources:
              limits:
                nvidia.com/gpu: "1"
                rdma/rdma_shared_device_eth: "1"
              requests:
                nvidia.com/gpu: "1"
                rdma/rdma_shared_device_eth: "1"
    Worker:
      replicas: 3
      restartPolicy: OnFailure
      template:
        metadata:
          annotations:
            k8s.v1.cni.cncf.io/networks: "example-net"
        spec:
          containers:
          - command:
            - /bin/bash
            - -c
            - "your container command"
            env:
            - name: NCCL_SOCKET_IFNAME
              value: "net1"
            - name: NCCL_IB_HCA
              value: "mlx5_1"
            image: quay.io/modh/training:py311-cuda121-torch241
            name: pytorch
            resources:
              limits:
                nvidia.com/gpu: "1"
                rdma/rdma_shared_device_eth: "1"
              requests:
                nvidia.com/gpu: "1"
                rdma/rdma_shared_device_eth: "1"

apiVersion: kubeflow.org/v1
kind: PyTorchJob
metadata:
  name: job
spec:
  pytorchReplicaSpecs:
    Master:
      replicas: 1
      restartPolicy: OnFailure
      template:
        metadata:
          annotations:
            k8s.v1.cni.cncf.io/networks: "example-net"
        spec:
          containers:
          - command:
            - /bin/bash
            - -c
            - "your container command"
            env:
            - name: NCCL_SOCKET_IFNAME
              value: "net1"
            - name: NCCL_IB_HCA
              value: "mlx5_1"
            image: quay.io/modh/training:py311-cuda121-torch241
            name: pytorch
            resources:
              limits:
                nvidia.com/gpu: "1"
                rdma/rdma_shared_device_eth: "1"
              requests:
                nvidia.com/gpu: "1"
                rdma/rdma_shared_device_eth: "1"
    Worker:
      replicas: 3
      restartPolicy: OnFailure
      template:
        metadata:
          annotations:
            k8s.v1.cni.cncf.io/networks: "example-net"
        spec:
          containers:
          - command:
            - /bin/bash
            - -c
            - "your container command"
            env:
            - name: NCCL_SOCKET_IFNAME
              value: "net1"
            - name: NCCL_IB_HCA
              value: "mlx5_1"
            image: quay.io/modh/training:py311-cuda121-torch241
            name: pytorch
            resources:
              limits:
                nvidia.com/gpu: "1"
                rdma/rdma_shared_device_eth: "1"
              requests:
                nvidia.com/gpu: "1"
                rdma/rdma_shared_device_eth: "1"