pcm-coordinator/desc/inference/inference.api

syntax = "v1"

import (
     "../schedule/pcm-schedule.api"
)

type (
     /******************image inference*************************/
     DeployInstance {
          Id string `json:"id"`
          DeployTaskId string `json:"deployTaskId"`
          InstanceId string `json:"instanceId"`
          InstanceName string `json:"instanceName"`
          AdapterId string `json:"adapterId"`
          AdapterName string `json:"adapterName"`
          ClusterId string `json:"clusterId"`
          ClusterName string `json:"clusterName"`
          ModelName string `json:"modelName"`
          ModelType string `json:"modelType"`
          InferCard string `json:"inferCard"`
          Status string `json:"status"`
          CreateTime string `json:"createTime"`
          UpdateTime string `json:"updateTime"`
          ClusterType string `json:"clusterType"`
     }

     /******************image inference*************************/
     ModelTypesResp {
          ModelTypes []string `json:"types"`
     }

     ModelNamesReq {
          Type string `form:"type"`
     }

     ModelNamesResp {
          ModelNames []string `json:"models"`
     }
     /******************image inference*************************/
     /******************instance center*************************/
     InstanceCenterReq{
          InstanceType    int32    `form:"instance_type"`
          InstanceClass   string   `form:"instance_class,optional"`
          InstanceName    string   `form:"instance_name,optional"`
     }
     InstanceCenterResp {
          InstanceCenterList []InstanceCenterList `json:"instanceCenterList" copier:"InstanceCenterList"`
          Code int32 `json:"code,omitempty"`
          Msg string `json:"msg,omitempty"`
          ErrorMsg string `json:"errorMsg,omitempty"`
          TotalCount int   `json:"totalCount"`
     }
     InstanceCenterList {
          LogoPath           string `json:"logo_path"`
          InstanceName       string `json:"instance_name"`
          InstanceType       int32 `json:"instance_type"`
          InstanceClass      string `json:"instance_class"`
          InstanceClassChinese      string `json:"instance_class_chinese"`
          Description        string `json:"description"`
          Version            string `json:"version"`
     }

     /******************instance center*************************/

     ImageInferenceReq {
          TaskName string `form:"taskName"`
          TaskDesc string `form:"taskDesc"`
          ModelType string `form:"modelType"`
          InstanceIds []int64 `form:"instanceIds"`
          Strategy string `form:"strategy,,optional"`
          StaticWeightMap map[string]map[string]int32 `form:"staticWeightMap,optional"`
          Replica     int32     `form:"replicas,optional"`
     }

     ImageInferenceResp {
          InferResults []*ImageResult `json:"result"`
     }

     ImageResult {
          ClusterId string `json:"clusterId"`
          ClusterName string `json:"clusterName"`
          ImageName string `json:"imageName"`
          Card string `json:"card"`
          ImageResult string `json:"imageResult"`
     }

     InferenceTaskDetailReq{
          taskId int64 `form:"taskId"`
     }

     InferenceTaskDetailResp{
          InferenceResults []InferenceResult `json:"data"`
          Code int32 `json:"code,omitempty"`
          Msg string `json:"msg,omitempty"`
     }

     InferenceResult{
          imageName string `json:"imageName"`
          TaskName string `json:"taskName"`
          TaskAiName string `json:"taskAiName"`
          result string `json:"result"`
          card string `json:"card"`
          clusterName string `json:"clusterName"`
     }

     /******************TextToText inference*************************/
     TextToTextInferenceReq{
          TaskName string `form:"taskName"`
          TaskDesc string `form:"taskDesc"`
          ModelType string `form:"modelType"`
          InstanceId int64 `form:"instanceId"`
     }
     TextToTextInferenceResp{

     }

     /******************TextToImage inference*************************/
     TextToImageInferenceReq{
          TaskName string `form:"taskName"`
          TaskDesc string `form:"taskDesc"`
          ModelName string `form:"modelName"`
          ModelType string `form:"modelType"`
          AiClusterIds []string `form:"aiClusterIds"`
     }
     TextToImageInferenceResp{
          Result []byte `json:"result"`
     }

     /******************Deploy instance*************************/
     GetDeployInstanceReq{
          AdapterId          string `form:"adapterId"`
          ClusterId      string `form:"clusterId"`
          Id      string `form:"id"`
          InstanceId string `form:"instanceId"`
     }

     GetDeployInstanceResp {
          Instance     interface{} `json:"instance"`
     }


     DeployInstanceListReq{
          PageInfo
     }

     DeployInstanceListResp {
          PageResult
     }

     StartDeployInstanceReq {
          AdapterId          string `form:"adapterId"`
          ClusterId      string `form:"clusterId"`
          Id      string `form:"id"`
          InstanceId string `form:"instanceId"`
     }

     StartDeployInstanceResp {
     }

     StopDeployInstanceReq {
          AdapterId          string `form:"adapterId"`
          ClusterId      string `form:"clusterId"`
          Id      string `form:"id"`
          InstanceId string `form:"instanceId"`
     }

     StopDeployInstanceResp {
     }

     DeployInstanceStatReq {

     }

     DeployInstanceStatResp {
          Running int32 `json:"running"`
          Total int32 `json:"total"`
     }

     InferenceTaskStatReq {}

     InferenceTaskStatResp {
          Running int32 `json:"running"`
          Total int32 `json:"total"`
     }

     StartAllByDeployTaskIdReq {
          Id      string `json:"deployTaskId"`
     }

     StartAllByDeployTaskIdResp {

     }

     StopAllByDeployTaskIdReq {
          Id      string `json:"deployTaskId"`
     }

     StopAllByDeployTaskIdResp {

     }

     GetRunningInstanceReq {
          Id      string `form:"deployTaskId"`
          AdapterId      string `form:"adapterId"`
     }
     GetRunningInstanceResp {
          List     interface{} `json:"list"`
     }

     GetDeployTasksByTypeReq {
          ModelType string `form:"modelType"`
     }

     GetDeployTasksByTypeResp {
          List     interface{} `json:"list"`
     }

     CreateDeployTaskReq {
          TaskName string `form:"taskName"`
          TaskDesc string `form:"taskDesc"`
          ModelName string `form:"modelName"`
          ModelType string `form:"modelType"`
          AdapterClusterMap map[string][]string `form:"adapterClusterMap"`
     }

     CreateDeployTaskResp {

     }

     GetAdaptersByModelReq {
          ModelName string `form:"modelName"`
          ModelType string `form:"modelType"`
     }

     GetAdaptersByModelResp {
          Adapters []*AdapterAvail `json:"adapters"`
     }

     AdapterAvail {
          AdapterId string `json:"adapterId"`
          AdapterName string `json:"adapterName"`
          Clusters []*ClusterAvail `json:"clusters"`
     }

     ClusterAvail {
          ClusterId string `json:"clusterId"`
          ClusterName string `json:"clusterName"`
     }

     // 推理任务
     CreateInferenceTaskReq {
          Name string `json:"name"`
          Description string `json:"description,optional"`
          Token string `json:"token,optional"`
          UserIp string `json:"userIp,optional"`
          JobResources JobResources `json:"jobResources"`
          DataDistributes DataDistribute `json:"dataDistributes"`
     }

     CreateInferenceTaskResp {
          TaskId string `json:"taskId"`
          TaskName string `json:"taskName"`
     }
)