services/ai_task_service/cluster/c2net.go

@@ -129,7 +129,7 @@ func convertNoteBookReq2Grampus(req entity.CreateNoteBookTaskRequest) (models.Cr
 	var commandGpuDebug = "mkdir -p /tmp/dataset;jupyter lab --ServerApp.shutdown_no_activity_timeout=%s --TerminalManager.cull_inactive_timeout=%s --TerminalManager.cull_interval=%s --MappingKernelManager.cull_idle_timeout=%s --MappingKernelManager.cull_interval=%s --MappingKernelManager.cull_connected=True --MappingKernelManager.cull_busy=True --no-browser --ip=0.0.0.0 --allow-root --notebook-dir='%s' --port=$OCTOPUS_NOTEBOOK_PORT --LabApp.token='' --LabApp.allow_origin='*' --LabApp.base_url=$OCTOPUS_NOTEBOOK_BASE_URL;"
 	command := fmt.Sprintf(commandGpuDebug, setting.CullIdleTimeout, setting.CullIdleTimeout, setting.CullInterval, setting.CullIdleTimeout, setting.CullInterval, codePath)
 	if models.DCU == req.Tasks[0].Spec.ComputeResource {
-		command = "cp -r /code /tmp;cp -r /dataset /tmp;cp -r /pretrainmodel /tmp;"
+		command = ""
 	}
 	if models.NPU == req.Tasks[0].Spec.ComputeResource {
 		command = ""

services/ai_task_service/container_builder/dataset_builder.go

@@ -26,22 +26,6 @@ func (b *DatasetBuilder) SetOpts(opts *entity.ContainerBuildOpts) {
 }
 
 func (b *DatasetBuilder) Build(ctx *context.CreationContext) ([]entity.ContainerData, *response.BizError) {
-	if b.Opts.Disable {
-		return nil, nil
-	}
-	uuid := ctx.Request.DatasetUUIDStr
-	if uuid == "" {
-		return nil, nil
-	}
-	datasetInfos, err := models.GetDatasetInfo4AITask(uuid)
-	if err != nil {
-		log.Error("GetDatasetInfo failed: %v", err)
-		return nil, response.DATASET_SELECT_ERROR
-	}
-	if len(datasetInfos) < len(strings.Split(uuid, ";")) {
-		log.Error("GetDatasetInfo count error.displayJobName=%s jobType=%s cluster=%s", ctx.Request.DisplayJobName, ctx.Request.JobType, ctx.Request.Cluster)
-		return nil, response.PARTIAL_DATASETS_NOT_AVAILABLE
-	}
 	var data []entity.ContainerData
 
 	//如果是智算GPU调试任务，需要把dataset文件夹也挂载，这样提交镜像时才不会把dataset下的文件提交到镜像中
@@ -76,6 +60,23 @@ func (b *DatasetBuilder) Build(ctx *context.CreationContext) ([]entity.Container
 		})
 	}
 
+	if b.Opts.Disable {
+		return data, nil
+	}
+	uuid := ctx.Request.DatasetUUIDStr
+	if uuid == "" {
+		return data, nil
+	}
+	datasetInfos, err := models.GetDatasetInfo4AITask(uuid)
+	if err != nil {
+		log.Error("GetDatasetInfo failed: %v", err)
+		return nil, response.DATASET_SELECT_ERROR
+	}
+	if len(datasetInfos) < len(strings.Split(uuid, ";")) {
+		log.Error("GetDatasetInfo count error.displayJobName=%s jobType=%s cluster=%s", ctx.Request.DisplayJobName, ctx.Request.JobType, ctx.Request.Cluster)
+		return nil, response.PARTIAL_DATASETS_NOT_AVAILABLE
+	}
+
 	for _, datasetInfo := range datasetInfos {
 		var name, objectKey, s3DownloadUrl string
 		//如果不是压缩包，那么文件名是去掉后缀以后的数据集名称

services/ai_task_service/container_builder/pre_model_builder.go

@@ -1,11 +1,10 @@
 package container_builder
 
 import (
+	"code.gitea.io/gitea/services/ai_model"
 	"path"
 	"strings"
 
-	"code.gitea.io/gitea/services/ai_model"
-
 	"code.gitea.io/gitea/routers/response"
 
 	"code.gitea.io/gitea/entity"
@@ -29,17 +28,39 @@ func (b *PretrainModelBuilder) SetOpts(opts *entity.ContainerBuildOpts) {
 }
 
 func (b *PretrainModelBuilder) Build(ctx *context.CreationContext) ([]entity.ContainerData, *response.BizError) {
+	form := ctx.Request
+	var preTrainModelEntity []entity.ContainerData
+	if ctx.Request.Cluster == entity.C2Net && (ctx.Request.JobType == models.JobTypeDebug || ctx.Request.JobType == models.JobTypeTrain) && ctx.Request.ComputeSource.Name == models.GPU {
+		//挂载一个文件夹保证容器内pretrainmodel目录提交镜像时不被打包
+		uploader := storage_helper.SelectStorageHelperFromStorageType(entity.OBS)
+		objectKey := path.Join(uploader.GetJobDefaultObjectKeyPrefix(form.JobName), "pretrain_model_mount")
+		uploader.MKDIR(objectKey, "pretrain model folder")
+		preTrainModelEntity = append(preTrainModelEntity, entity.ContainerData{
+			Name:          "pretrain_model_mount",
+			Bucket:        uploader.GetBucket(),
+			EndPoint:      uploader.GetEndpoint(),
+			ObjectKey:     objectKey + "/",
+			ReadOnly:      false,
+			ContainerPath: b.Opts.ContainerPath,
+			RealPath:      uploader.GetRealPath(objectKey),
+			S3DownloadUrl: uploader.GetS3DownloadUrl(objectKey),
+			IsDir:         true,
+			IsOverwrite:   true,
+			IsNeedUnzip:   false,
+		})
+	}
+
 	if b.Opts.Disable {
-		return nil, nil
+		return preTrainModelEntity, nil
 	}
-	form := ctx.Request
+
 	storageTypes := b.Opts.AcceptStorageType
 	if storageTypes == nil || len(storageTypes) == 0 {
 		return nil, response.SYSTEM_ERROR
 	}
 	//未选择预训练模型，跳过此步
 	if form.PretrainModelId == "" {
-		return nil, nil
+		return preTrainModelEntity, nil
 	}
 	//查出模型数据
 	uuids := strings.Split(form.PretrainModelId, ";")
@@ -48,7 +69,7 @@ func (b *PretrainModelBuilder) Build(ctx *context.CreationContext) ([]entity.Con
 		log.Error("Can not find model", err)
 		return nil, response.MODEL_NOT_EXISTS
 	}
-	var preTrainModelEntity []entity.ContainerData
+
 	for _, m := range modelInfoMaps {
 		ai_model.InitModelMeta(m.ID)
 		data, err := b.buildModelData(m, form.JobName)
@@ -57,6 +78,7 @@ func (b *PretrainModelBuilder) Build(ctx *context.CreationContext) ([]entity.Con
 		}
 		preTrainModelEntity = append(preTrainModelEntity, data)
 	}
+
 	return preTrainModelEntity, nil
 }
 
@@ -101,7 +123,7 @@ func (b *PretrainModelBuilder) buildModelData(m *models.AiModelManage, jobName s
 		Bucket:        uploader.GetBucket(),
 		EndPoint:      uploader.GetEndpoint(),
 		ObjectKey:     preTrainModelPath,
-		ReadOnly:      b.Opts.ReadOnly,
+		ReadOnly:      false,
 		ContainerPath: path.Join(b.Opts.ContainerPath, m.Name),
 		RealPath:      uploader.GetRealPath(preTrainModelPath),
 		S3DownloadUrl: uploader.GetS3DownloadUrl(preTrainModelPath),