aiforge/modules/grampus/grampus.go

package grampus

import (
	"code.gitea.io/gitea/models"
	"code.gitea.io/gitea/modules/context"
	"code.gitea.io/gitea/modules/log"
	"code.gitea.io/gitea/modules/notification"
	"code.gitea.io/gitea/modules/timeutil"
	"strings"
)

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）"

	CodePath         = "/code/"
	OutputPath       = "/output/"
	ResultPath       = "/result/"
	LogPath          = "/log/"
	JobPath          = "/job/"
	OrderDesc        = "desc" //向下查询
	OrderAsc         = "asc"  //向上查询
	Lines            = 500
	TrainUrl         = "train_url"
	DataUrl          = "data_url"
	ResultUrl        = "result_url"
	CkptUrl          = "ckpt_url"
	DeviceTarget     = "device_target"
	Ascend           = "Ascend"
	PerPage          = 10
	IsLatestVersion  = "1"
	NotLatestVersion = "0"
	VersionCount     = 1

	SortByCreateTime  = "create_time"
	ConfigTypeCustom  = "custom"
	TotalVersionCount = 1

	ProcessorTypeNPU = "npu.huawei.com/NPU"
	ProcessorTypeGPU = "nvidia.com/gpu"
)

var (
	poolInfos   *models.PoolInfos
	FlavorInfos *models.FlavorInfos
	ImageInfos  *models.ImageInfosModelArts
)

type GenerateTrainJobReq struct {
	JobName        string
	Command        string
	ResourceSpecId string
	ImageUrl       string //与image_id二选一，都有的情况下优先image_url
	ImageId        string

	DisplayJobName    string
	Uuid              string
	Description       string
	CodeObsPath       string
	BootFile          string
	BootFileUrl       string
	DataUrl           string
	TrainUrl          string
	WorkServerNumber  int
	EngineID          int64
	CommitID          string
	IsLatestVersion   string
	BranchName        string
	PreVersionId      int64
	PreVersionName    string
	FlavorName        string
	VersionCount      int
	EngineName        string
	TotalVersionCount int
	ComputeResource   string
	DatasetName       string
	Params            string
}

func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error) {
	createTime := timeutil.TimeStampNow()
	jobResult, err := createJob(models.CreateGrampusJobRequest{
		Name: req.JobName,
		Tasks: []models.GrampusTasks{
			{
				Name:           req.JobName,
				Command:        req.Command,
				ResourceSpecId: req.ResourceSpecId,
				ImageId:        req.ImageId,
				ImageUrl:       req.ImageUrl,
			},
		},
	})
	if err != nil {
		log.Error("createJob failed: %v", err.Error())
		return err
	}

	jobID := jobResult.JobInfo.JobID
	err = models.CreateCloudbrain(&models.Cloudbrain{
		Status:            TransTrainJobStatus(jobResult.JobInfo.Status),
		UserID:            ctx.User.ID,
		RepoID:            ctx.Repo.Repository.ID,
		JobID:             jobID,
		JobName:           req.JobName,
		DisplayJobName:    req.DisplayJobName,
		JobType:           string(models.JobTypeTrain),
		Type:              models.TypeCloudBrainGrampus,
		Uuid:              req.Uuid,
		DatasetName:       req.DatasetName,
		CommitID:          req.CommitID,
		IsLatestVersion:   req.IsLatestVersion,
		ComputeResource:   req.ComputeResource,
		ImageID:           req.ImageId,
		TrainUrl:          req.TrainUrl,
		BranchName:        req.BranchName,
		Parameters:        req.Params,
		BootFile:          req.BootFile,
		DataUrl:           req.DataUrl,
		FlavorCode:        req.ResourceSpecId,
		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
	}

	var actionType models.ActionType
	if req.ComputeResource == models.NPUResource {
		actionType = models.ActionCreateTrainTask
	} else if req.ComputeResource == models.GPUResource {
		actionType = models.ActionCreateGPUTrainTask
	}
	notification.NotifyOtherTask(ctx.User, ctx.Repo.Repository, jobID, req.DisplayJobName, actionType)

	return nil
}

func TransTrainJobStatus(status string) string {
	if status == "pending" {
		status = "waiting"
	}

	return strings.ToUpper(status)
}