diff --git a/README-CN.md b/README-CN.md index c4ca4647..ab9563b2 100644 --- a/README-CN.md +++ b/README-CN.md @@ -438,6 +438,7 @@ ROS 模板的示例和最佳实践。模板分类如下： | [pai-lingjun-serverless-LLM-best-practice.yml](documents/solution/ai/upai-lingjun-serverless-LLM-best-practice.yml) | PAI灵骏智算资源（Serverless版）大模型最佳实践。 | | [use-gpu-ecs-to-deploy-chatGLM.yaml](documents/solution/ai/use-gpu-ecs-to-deploy-chatGLM.yaml) | 向量数据库构建企业智能知识库。 ｜ [解决方案](https://aliyun.com/solution/tech-solution/baeeikb) | | [build-a-dialogue-model-based-on-ChatGLM-and-LangChain.yml](documents/solution/ai/build-a-dialogue-model-based-on-ChatGLM-and-LangChain.yml) | 基于ChatGLM和LangChain搭建对话模型。 | +| [ack-ai-fine-tuning.yml](documents/solution/ai/ack-ai-fine-tuning.yml) | 使用云原生AI套件提交模型微调训练任务与部署GPU共享推理服务。 | - data-analysis diff --git a/README.md b/README.md index 4a9682c3..f50a2e11 100644 --- a/README.md +++ b/README.md @@ -440,6 +440,7 @@ Examples and best practices of ROS templates. The templates are categorized as f | [use-pai-to-deploy-stable-diffusion-for-AI-painting.yml](documents/solution/ai/use-pai-to-deploy-stable-diffusion-for-AI-painting.yml) | How to quickly start Stable Diffusion on Alibaba Cloud and easily play with AI painting. | | [use-gpu-ecs-to-deploy-chatGLM.yaml](documents/solution/ai/use-gpu-ecs-to-deploy-chatGLM.yaml) | Large model combines the AnalyticDB to build Chatbot. | | [build-a-dialogue-model-based-on-ChatGLM-and-LangChain.yml](documents/solution/ai/build-a-dialogue-model-based-on-ChatGLM-and-LangChain.yml) | Build a dialogue model based on ChatGLM and LangChain. | +| [ack-ai-fine-tuning.yml](documents/solution/ai/ack-ai-fine-tuning.yml) | Use the cloud-native AI suite to submit model fine-tuning training tasks and deploy GPU shared inference services. | diff --git a/documents/solution/ai/ack-ai-fine-tuning.yml b/documents/solution/ai/ack-ai-fine-tuning.yml new file mode 100644 index 00000000..b2de79a1 --- /dev/null +++ b/documents/solution/ai/ack-ai-fine-tuning.yml @@ -0,0 +1,220 @@ +ROSTemplateFormatVersion: '2015-09-01' +Description: + en: Use the cloud-native AI suite to submit model fine-tuning training tasks and deploy GPU shared inference services. + zh-cn: 使用云原生AI套件提交模型微调训练任务与部署GPU共享推理服务。 +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（-）。 + Default: ai-test + AllowedPattern: ^[a-zA-Z0-9\u4e00-\u9fa5][-a-zA-Z0-9\u4e00-\u9fa5]{0,62}$ + 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.
+ It must consist three of the the following character types: uppercase letters, lowercase letters, digits, and special characters.
+ Special characters include ()`~!@#$%^&*_-+=|{}[]:;'<>,.?/.
+ zh-cn: 长度为8-30位，需包含大写字母、小写字母、特殊符号和数字中的三个，允许的特殊字符包括()`~!@#$%^&*_-+=|{}[]:;'<>,.?/。 + AssociationProperty: ALIYUN::ECS::Instance::Password + NoEcho: true + 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 + SecurityGroupIngress: + - PortRange: 22/22 + Priority: 1 + SourceCidrIp: 0.0.0.0/0 + IpProtocol: tcp + NicType: internet + 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 + NatGateway: + Type: ALIYUN::VPC::NatGateway + Properties: + ZoneId: + Ref: ZoneId + VpcId: + Ref: Vpc + VSwitchId: + Ref: Vswitch + NatGatewayName: + Fn::Sub: ${CommonName}-ngw + 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.gn7i-c16g1.4xlarge + WorkerSystemDiskCategory: cloud_essd + WorkerSystemDiskSize: 120 + LoginPassword: + Ref: LoginPassword + Runtime: + Name: containerd + Version: 1.6.20 + CloudMonitorFlags: false + ZoneIds: + - Ref: ZoneId + DependsOn: + - ModuleAcsCsProvision + - NatGateway + Nas: + Type: ALIYUN::NAS::FileSystem + Properties: + ProtocolType: NFS + FileSystemType: standard + StorageType: Capacity + ZoneId: + Ref: ZoneId + VpcId: + Ref: Vpc + VSwitchId: + Ref: Vswitch + Description: + Fn::Sub: ${CommonName}-nas + NasMountTarget: + Type: ALIYUN::NAS::MountTarget + Properties: + VpcId: + Ref: Vpc + VSwitchId: + Ref: Vswitch + Status: Active + FileSystemId: + Ref: Nas + NetworkType: Vpc + AccessGroupName: DEFAULT_VPC_GROUP_NAME +Outputs: + EcsLoginAddress: + Description: + zh-cn: ECS登陆地址。 + en: Ecs login address. + Value: + Fn::Sub: + - https://ecs-workbench.aliyun.com/?from=EcsConsole&instanceType=ecs®ionId=${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 + Hidden: + - CommonName + TemplateTags: + - acs:technical-solution:ack:训练大模型及部署GPU共享推理服务