提交代码。

Signed-off-by: zouap <zouap@pcl.ac.cn>

models/cloudbrain.go

@@ -164,12 +164,12 @@ type Cloudbrain struct {
 	FlavorName        string //规格名称
 	EngineName        string //引擎名称
 	TotalVersionCount int    //任务的所有版本数量，包括删除的
-
-	LabelName    string //标签名称
-	ModelName    string //模型名称
-	ModelVersion string //模型版本
-	CkptName     string //权重文件名称
-	ResultUrl    string //推理结果的obs路径
+	UserImageCommand  string `xorm:"varchar(1000)"` //描述
+	LabelName         string //标签名称
+	ModelName         string //模型名称
+	ModelVersion      string //模型版本
+	CkptName          string //权重文件名称
+	ResultUrl         string //推理结果的obs路径
 
 	User                  *User       `xorm:"-"`
 	Repo                  *Repository `xorm:"-"`

modules/modelarts/modelarts.go

@@ -19,14 +19,14 @@ import (
 
 const (
 	//notebook
-	storageTypeOBS     = "obs"
-	autoStopDuration   = 4 * 60 * 60
-	autoStopDurationMs = 4 * 60 * 60 * 1000
-
-	DataSetMountPath = "/home/ma-user/work"
-	NotebookEnv      = "Python3"
-	NotebookType     = "Ascend"
-	FlavorInfo       = "Ascend: 1*Ascend 910 CPU: 24 核 96GiB （modelarts.kat1.xlarge）"
+	storageTypeOBS                 = "obs"
+	autoStopDuration               = 4 * 60 * 60
+	autoStopDurationMs             = 4 * 60 * 60 * 1000
+	MORDELART_USER_IMAGE_ENGINE_ID = 1000000
+	DataSetMountPath               = "/home/ma-user/work"
+	NotebookEnv                    = "Python3"
+	NotebookType                   = "Ascend"
+	FlavorInfo                     = "Ascend: 1*Ascend 910 CPU: 24 核 96GiB （modelarts.kat1.xlarge）"
 
 	//train-job
 	// ResourcePools  = "{\"resource_pool\":[{\"id\":\"pool1328035d\", \"value\":\"专属资源池\"}]}"
@@ -152,6 +152,7 @@ type Engine struct {
 	Info []struct {
 		ID    int    `json:"id"`
 		Value string `json:"value"`
+		Url   string `json:"url"`
 	} `json:"engine"`
 }
 
@@ -314,6 +315,78 @@ func GenerateNotebook2(ctx *context.Context, displayJobName, jobName, uuid, desc
 	return nil
 }
 
+func GenerateTrainJobByUserImage(ctx *context.Context, req *GenerateTrainJobReq) (err error) {
+	createTime := timeutil.TimeStampNow()
+	jobResult, err := createTrainJobUserImage(models.CreateUserImageTrainJobParams{
+		JobName:     req.JobName,
+		Description: req.Description,
+		Config: models.UserImageConfig{
+			WorkServerNum: req.WorkServerNumber,
+			AppUrl:        req.CodeObsPath,
+			BootFileUrl:   req.BootFileUrl,
+			DataUrl:       req.DataUrl,
+			TrainUrl:      req.TrainUrl,
+			LogUrl:        req.LogUrl,
+			PoolID:        req.PoolID,
+			CreateVersion: true,
+			Flavor: models.Flavor{
+				Code: req.FlavorCode,
+			},
+			Parameter:    req.Parameters,
+			UserImageUrl: req.UserImageUrl,
+			UserCommand:  req.UserCommand,
+		},
+	})
+	if err != nil {
+		log.Error("CreateJob failed: %v", err.Error())
+		return err
+	}
+
+	jobId := strconv.FormatInt(jobResult.JobID, 10)
+	err = models.CreateCloudbrain(&models.Cloudbrain{
+		Status:            TransTrainJobStatus(jobResult.Status),
+		UserID:            ctx.User.ID,
+		RepoID:            ctx.Repo.Repository.ID,
+		JobID:             jobId,
+		JobName:           req.JobName,
+		DisplayJobName:    req.DisplayJobName,
+		JobType:           string(models.JobTypeTrain),
+		Type:              models.TypeCloudBrainTwo,
+		VersionID:         jobResult.VersionID,
+		VersionName:       jobResult.VersionName,
+		Uuid:              req.Uuid,
+		DatasetName:       req.DatasetName,
+		CommitID:          req.CommitID,
+		IsLatestVersion:   req.IsLatestVersion,
+		ComputeResource:   models.NPUResource,
+		EngineID:          MORDELART_USER_IMAGE_ENGINE_ID,
+		Image:             req.UserImageUrl,
+		UserImageCommand:  req.UserCommand,
+		TrainUrl:          req.TrainUrl,
+		BranchName:        req.BranchName,
+		Parameters:        req.Params,
+		BootFile:          req.BootFile,
+		DataUrl:           req.DataUrl,
+		LogUrl:            req.LogUrl,
+		FlavorCode:        req.FlavorCode,
+		Description:       req.Description,
+		WorkServerNumber:  req.WorkServerNumber,
+		FlavorName:        req.FlavorName,
+		EngineName:        req.EngineName,
+		VersionCount:      req.VersionCount,
+		TotalVersionCount: req.TotalVersionCount,
+		CreatedUnix:       createTime,
+		UpdatedUnix:       createTime,
+	})
+
+	if err != nil {
+		log.Error("CreateCloudbrain(%s) failed:%v", req.DisplayJobName, err.Error())
+		return err
+	}
+	notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, jobId, req.DisplayJobName, models.ActionCreateTrainTask)
+	return nil
+}
+
 func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error) {
 	createTime := timeutil.TimeStampNow()
 	jobResult, err := createTrainJob(models.CreateTrainJobParams{
@@ -497,6 +570,100 @@ func GenerateTrainJobVersion(ctx *context.Context, req *GenerateTrainJobReq, job
 	return err
 }
 
+func GenerateTrainJobVersionByUserImage(ctx *context.Context, req *GenerateTrainJobReq, jobId string) (err error) {
+	createTime := timeutil.TimeStampNow()
+	jobResult, err := createTrainJobUserImage(models.CreateUserImageTrainJobParams{
+		JobName:     req.JobName,
+		Description: req.Description,
+		Config: models.UserImageConfig{
+			WorkServerNum: req.WorkServerNumber,
+			AppUrl:        req.CodeObsPath,
+			BootFileUrl:   req.BootFileUrl,
+			DataUrl:       req.DataUrl,
+			TrainUrl:      req.TrainUrl,
+			LogUrl:        req.LogUrl,
+			PoolID:        req.PoolID,
+			CreateVersion: true,
+			Flavor: models.Flavor{
+				Code: req.FlavorCode,
+			},
+			Parameter:    req.Parameters,
+			UserImageUrl: req.UserImageUrl,
+			UserCommand:  req.UserCommand,
+		},
+	})
+	if err != nil {
+		log.Error("CreateJob failed: %v", err.Error())
+		return err
+	}
+
+	var jobTypes []string
+	jobTypes = append(jobTypes, string(models.JobTypeTrain))
+	repo := ctx.Repo.Repository
+	VersionTaskList, VersionListCount, err := models.CloudbrainsVersionList(&models.CloudbrainsOptions{
+		RepoID:   repo.ID,
+		Type:     models.TypeCloudBrainTwo,
+		JobTypes: jobTypes,
+		JobID:    strconv.FormatInt(jobResult.JobID, 10),
+	})
+	if err != nil {
+		ctx.ServerError("Cloudbrain", err)
+		return err
+	}
+	//将当前版本的isLatestVersion设置为"1"和任务数量更新,任务数量包括当前版本数VersionCount和历史创建的总版本数TotalVersionCount
+
+	err = models.CreateCloudbrain(&models.Cloudbrain{
+		Status:            TransTrainJobStatus(jobResult.Status),
+		UserID:            ctx.User.ID,
+		RepoID:            ctx.Repo.Repository.ID,
+		JobID:             strconv.FormatInt(jobResult.JobID, 10),
+		JobName:           req.JobName,
+		DisplayJobName:    req.DisplayJobName,
+		JobType:           string(models.JobTypeTrain),
+		Type:              models.TypeCloudBrainTwo,
+		VersionID:         jobResult.VersionID,
+		VersionName:       jobResult.VersionName,
+		Uuid:              req.Uuid,
+		DatasetName:       req.DatasetName,
+		CommitID:          req.CommitID,
+		IsLatestVersion:   req.IsLatestVersion,
+		PreVersionName:    req.PreVersionName,
+		ComputeResource:   models.NPUResource,
+		EngineID:          MORDELART_USER_IMAGE_ENGINE_ID,
+		Image:             req.UserImageUrl,
+		UserImageCommand:  req.UserCommand,
+		TrainUrl:          req.TrainUrl,
+		BranchName:        req.BranchName,
+		Parameters:        req.Params,
+		BootFile:          req.BootFile,
+		DataUrl:           req.DataUrl,
+		LogUrl:            req.LogUrl,
+		PreVersionId:      req.PreVersionId,
+		FlavorCode:        req.FlavorCode,
+		Description:       req.Description,
+		WorkServerNumber:  req.WorkServerNumber,
+		FlavorName:        req.FlavorName,
+		EngineName:        req.EngineName,
+		TotalVersionCount: VersionTaskList[0].TotalVersionCount + 1,
+		VersionCount:      VersionListCount + 1,
+		CreatedUnix:       createTime,
+		UpdatedUnix:       createTime,
+	})
+	if err != nil {
+		log.Error("CreateCloudbrain(%s) failed:%v", req.JobName, err.Error())
+		return err
+	}
+
+	//将训练任务的上一版本的isLatestVersion设置为"0"
+	err = models.SetVersionCountAndLatestVersion(strconv.FormatInt(jobResult.JobID, 10), VersionTaskList[0].VersionName, VersionCount, NotLatestVersion, TotalVersionCount)
+	if err != nil {
+		ctx.ServerError("Update IsLatestVersion failed", err)
+		return err
+	}
+
+	return err
+}
+
 func TransTrainJobStatus(status int) string {
 	switch status {
 	case 0:

routers/repo/ai_model_convert.go

@@ -39,12 +39,12 @@ const (
 	Success                    = "S000"
 	GPU_PYTORCH_IMAGE          = "dockerhub.pcl.ac.cn:5000/user-images/openi:tensorRT_7_zouap"
 	GPU_TENSORFLOW_IMAGE       = "dockerhub.pcl.ac.cn:5000/user-images/openi:tf2onnx"
-
-	PytorchOnnxBootFile   = "convert_pytorch.py"
-	PytorchTrTBootFile    = "convert_pytorch_tensorrt.py"
-	MindsporeBootFile     = "convert_mindspore.py"
-	TensorFlowNpuBootFile = "convert_tensorflow.py"
-	TensorFlowGpuBootFile = "convert_tensorflow_gpu.py"
+	NPU_MINDSPORE_16_IMAGE     = "swr.cn-south-222.ai.pcl.cn/openi/mindspore1.6.1_train_v1_openi:v3_ascend"
+	PytorchOnnxBootFile        = "convert_pytorch.py"
+	PytorchTrTBootFile         = "convert_pytorch_tensorrt.py"
+	MindsporeBootFile          = "convert_mindspore.py"
+	TensorFlowNpuBootFile      = "convert_tensorflow.py"
+	TensorFlowGpuBootFile      = "convert_tensorflow_gpu.py"
 
 	ConvertRepoPath = "https://git.openi.org.cn/zouap/npu_test"
 	REPO_ID         = 33267
@@ -227,7 +227,7 @@ func createNpuTrainJob(modelConvert *models.AiModelConvert, ctx *context.Context
 		PoolID:           NPU_PoolID,
 		//Parameters:       param,
 		BranchName:   DefaultBranchName,
-		UserImageUrl: "swr.cn-south-222.ai.pcl.cn/openi/mindspore1.6.1_train_v1_openi:v3_ascend",
+		UserImageUrl: NPU_MINDSPORE_16_IMAGE,
 		UserCommand:  userCommand,
 	}
 	result, err := modelarts.GenerateModelConvertTrainJob(req)