ack-ai-fine-tuning.yml

ROSTemplateFormatVersion: '2015-09-01'
Description:
  zh-cn: 创建Kubernetes集群，配置VPC、安全组、NAS，部署模型微调训练任务与GPU共享推理服务。
  en: Create a Kubernetes cluster, configure Virtual Private Cloud (VPC), Security
    Groups, and Network Attached Storage (NAS), and deploy model fine-tuning training
    tasks along with GPU-shared inference services.
Parameters:
  AckName:
    Type: String
    Label:
      en: Cluster Name
      zh-cn: 集群名称
    Description:
      en: The name must be 1 to 63 characters in length and can contain letters, Chinese
        characters, digits, and hyphens (-).
      zh-cn: 名称为1~63个字符，可包含数字、汉字、英文字符或中划线（-）。
    AllowedPattern: ^[a-zA-Z0-9\u4e00-\u9fa5][-a-zA-Z0-9\u4e00-\u9fa5]{0,62}$
    AssociationProperty: AutoCompleteInput
    AssociationPropertyMetadata:
      Length: 5
      Prefix: ai-test-
      CharacterClasses:
        - Class: lowercase
          min: 1
  ZoneId:
    Type: String
    Label:
      en: VSwitch Available Zone
      zh-cn: 可用区
    AssociationProperty: ALIYUN::VPC::Zone::ZoneId
    Description:
      en: If the available zone cannot be selected, please switch regions.
      zh-cn: 如果选择不到可用区，请切换地域。
  LoginPassword:
    Type: String
    Label:
      en: Set node login password
      zh-cn: 设置节点登录密码
    Description:
      en: |-
        The password must be 8 to 32 characters in length. <br>
        It must consist three of the the following character types: uppercase letters, lowercase letters, digits, and special characters. <br>
        Special characters include <span style="background:#E7E9EB;"><b>()`~!@#$%^&*_-+=|{}[]:;'<>,.?/</b></span>.<br>
      zh-cn: 长度为8-30位，需包含大写字母、小写字母、特殊符号和数字中的三个，允许的特殊字符包括<span style="background:#E7E9EB;"><b>()`~!@#$%^&*_-+=|{}[]:;'<>,.?/</b></span>。
    AssociationProperty: ALIYUN::ECS::Instance::Password
    NoEcho: true
  NasZoneId:
    Type: String
    Label:
      en: NAS Available Zone
      zh-cn: NAS 可用区
    AssociationProperty: ALIYUN::VPC::Zone::ZoneId
  CommonName:
    Type: String
    Default: ack-ai
Resources:
  ModuleAcsCsProvision:
    Type: MODULE::ACS::CS::Provision
    Version: default
  Vpc:
    Type: ALIYUN::ECS::VPC
    Properties:
      CidrBlock: 192.168.0.0/16
      VpcName:
        Fn::Sub: ${CommonName}-vpc
  Vswitch:
    Type: ALIYUN::ECS::VSwitch
    Properties:
      ZoneId:
        Ref: ZoneId
      VpcId:
        Ref: Vpc
      CidrBlock: 192.168.0.0/24
      VSwitchName:
        Fn::Sub: ${CommonName}-${ZoneId}-vsw
  SecurityGroup:
    Type: ALIYUN::ECS::SecurityGroup
    Properties:
      VpcId:
        Ref: Vpc
      SecurityGroupName:
        Fn::Sub: ${CommonName}-sg
      SecurityGroupEgress:
        - PortRange: -1/-1
          Priority: 1
          IpProtocol: all
          DestCidrIp: 0.0.0.0/0
          NicType: internet
        - PortRange: -1/-1
          Priority: 1
          IpProtocol: all
          DestCidrIp: 0.0.0.0/0
          NicType: intranet
  Eip:
    Type: ALIYUN::VPC::EIP
    Properties:
      Name:
        Fn::Sub: ${CommonName}-eip
      Bandwidth: 100
      InternetChargeType: PayByTraffic
  EipAssociation:
    Type: ALIYUN::VPC::EIPAssociation
    DependsOn: Sleep
    Properties:
      AllocationId:
        Ref: Eip
      InstanceId:
        Fn::Jq:
          - First
          - .[0].instance_id
          - Fn::GetAtt:
              - Ack
              - Nodes
  Sleep:
    Type: ALIYUN::ROS::Sleep
    DependsOn: Ack
    Properties:
      CreateDuration: 180
  Ack:
    Type: ALIYUN::CS::ManagedKubernetesCluster
    Properties:
      VpcId:
        Ref: Vpc
      SecurityGroupId:
        Ref: SecurityGroup
      VSwitchIds:
        - Ref: Vswitch
      Name:
        Ref: AckName
      NumOfNodes: 1
      ProxyMode: ipvs
      ClusterSpec: ack.pro.small
      ServiceCidr: 172.16.0.0/16
      ContainerCidr: 10.0.0.0/8
      NodeCidrMask: 26
      Addons:
        - Name: flannel
        - Name: csi-plugin
        - Name: csi-provisioner
        - Name: storage-operator
          Config: '{"CnfsOssEnable":"false","CnfsNasEnable":"false"}'
        - Name: nginx-ingress-controller
          Config: '{"IngressSlbNetworkType":"internet","IngressSlbSpec":"slb.s2.small"}'
        - Name: ack-node-local-dns
        - Name: arms-prometheus
      EndpointPublicAccess: true
      SnatEntry: true
      IsEnterpriseSecurityGroup: true
      WorkerInstanceTypes:
        - ecs.gn6i-c24g1.12xlarge
        - ecs.gn6v-c8g1.4xlarge
        - ecs.gn6i-c24g1.24xlarge
      WorkerSystemDiskCategory: cloud_essd
      WorkerSystemDiskSize: 120
      LoginPassword:
        Ref: LoginPassword
      Runtime:
        Name: containerd
        Version: 1.6.20
      CloudMonitorFlags: false
      ZoneIds:
        - Ref: ZoneId
    DependsOn:
      - ModuleAcsCsProvision
  Nas:
    Type: ALIYUN::NAS::FileSystem
    Properties:
      ProtocolType: NFS
      FileSystemType: standard
      StorageType: Capacity
      ZoneId:
        Ref: NasZoneId
      VpcId:
        Ref: Vpc
      VSwitchId:
        Ref: Vswitch
      Description:
        Fn::Sub: ${CommonName}-nas
  NasMountTarget:
    Type: ALIYUN::NAS::MountTarget
    Properties:
      VpcId:
        Ref: Vpc
      VSwitchId:
        Ref: Vswitch
      Status: Active
      FileSystemId:
        Ref: Nas
      NetworkType: Vpc
      AccessGroupName: DEFAULT_VPC_GROUP_NAME
Outputs:
  EcsLoginAddress:
    Description:
      zh-cn: ECS登录地址。
      en: Ecs login address.
    Value:
      Fn::Sub:
        - https://ecs-workbench.aliyun.com/?from=EcsConsole&instanceType=ecs&regionId=${Region}&instanceId=${InstanceId}
        - InstanceId:
            Fn::Jq:
              - First
              - .[0].instance_id
              - Fn::GetAtt:
                  - Ack
                  - Nodes
          Region:
            Ref: ALIYUN::Region
Metadata:
  ALIYUN::ROS::Interface:
    ParameterGroups:
      - Parameters:
        - AckName
        - ZoneId
        - LoginPassword
        - NasZoneId
    Hidden:
      - CommonName
    TemplateTags:
      - acs:technical-solution:ack:一键训练大模型及部署GPU共享推理服务-tech_solu_82