Merge pull request '智算网络NPU训练任务优化--取消代码下载时对外网的依赖' (#3029) from npu-internet-limit into V20221019

Reviewed-on: https://git.openi.org.cn/OpenI/aiforge/pulls/3029
Reviewed-by: ychao_1983 <ychao_1983@sina.com>

models/cloudbrain.go

@@ -1490,6 +1490,8 @@ type GrampusTasks struct {
 	ReplicaNum     int              `json:"replicaNum"`
 	Datasets       []GrampusDataset `json:"datasets"`
 	Models         []GrampusDataset `json:"models"`
+	Code           GrampusDataset   `json:"code"`
+	BootFile       string           `json:"bootFile"`
 }
 
 type GrampusDataset struct {

modules/grampus/grampus.go

@@ -1,6 +1,7 @@
 package grampus
 
 import (
+	"code.gitea.io/gitea/modules/cloudbrain"
 	"encoding/json"
 	"strings"
 
@@ -73,6 +74,7 @@ type GenerateTrainJobReq struct {
 	PreTrainModelPath string
 	PreTrainModelUrl  string
 	Spec              *models.Specification
+	CodeName          string
 }
 
 func getEndPoint() string {
@@ -102,6 +104,7 @@ func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error
 	centerID, centerName := getCentersParamter(ctx, req)
 
 	var datasetGrampus, modelGrampus []models.GrampusDataset
+	var codeGrampus models.GrampusDataset
 	if ProcessorTypeNPU == req.ProcessType {
 		datasetGrampus = getDatasetGrampus(req.DatasetInfos)
 		if len(req.ModelName) != 0 {
@@ -114,6 +117,12 @@ func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error
 				},
 			}
 		}
+		codeGrampus = models.GrampusDataset{
+			Name:      req.CodeName,
+			Bucket:    setting.Bucket,
+			EndPoint:  getEndPoint(),
+			ObjectKey: req.CodeObsPath + cloudbrain.DefaultBranchName + ".zip",
+		}
 	}
 
 	jobResult, err := createJob(models.CreateGrampusJobRequest{
@@ -130,6 +139,8 @@ func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error
 				ReplicaNum:     1,
 				Datasets:       datasetGrampus,
 				Models:         modelGrampus,
+				Code:           codeGrampus,
+				BootFile:       req.BootFile,
 			},
 		},
 	})

routers/repo/grampus.go

@@ -713,6 +713,7 @@ func grampusTrainJobNpuCreate(ctx *context.Context, form auth.CreateGrampusTrain
 		DatasetNames:      datasetNames,
 		DatasetInfos:      datasetInfos,
 		Spec:              spec,
+		CodeName:          strings.ToLower(repo.Name),
 	}
 	if form.ModelName != "" { //使用预训练模型训练
 		req.ModelName = form.ModelName
@@ -977,8 +978,7 @@ func generateCommand(repoName, processorType, codeRemotePath, dataRemotePath, bo
 	command += "pwd;cd " + workDir + fmt.Sprintf(grampus.CommandPrepareScript, setting.Grampus.SyncScriptProject, setting.Grampus.SyncScriptProject)
 	//download code & dataset
 	if processorType == grampus.ProcessorTypeNPU {
-		commandDownload := "./downloader_for_obs " + setting.Bucket + " " + codeRemotePath + " " + grampus.CodeArchiveName + ";"
-		command += commandDownload
+		//no need to download code & dataset by internet
 	} else if processorType == grampus.ProcessorTypeGPU {
 		commandDownload := "./downloader_for_minio " + setting.Grampus.Env + " " + codeRemotePath + " " + grampus.CodeArchiveName + " '" + dataRemotePath + "' '" + datasetName + "'"
 		commandDownload = processPretrainModelParameter(pretrainModelPath, pretrainModelFileName, commandDownload)
@@ -987,8 +987,7 @@ func generateCommand(repoName, processorType, codeRemotePath, dataRemotePath, bo
 
 	//unzip code & dataset
 	if processorType == grampus.ProcessorTypeNPU {
-		commandUnzip := "cd " + workDir + "code;unzip -q master.zip;"
-		command += commandUnzip
+		//no need to process
 	} else if processorType == grampus.ProcessorTypeGPU {
 		unZipDatasetCommand := generateDatasetUnzipCommand(datasetName)
 		commandUnzip := "cd " + workDir + "code;unzip -q master.zip;echo \"start to unzip dataset\";cd " + workDir + "dataset;" + unZipDatasetCommand
@@ -1025,7 +1024,7 @@ func generateCommand(repoName, processorType, codeRemotePath, dataRemotePath, bo
 
 	var commandCode string
 	if processorType == grampus.ProcessorTypeNPU {
-		commandCode = "/bin/bash /home/work/run_train_for_openi.sh " + workDir + "code/" + strings.ToLower(repoName) + "/" + bootFile + " /tmp/log/train.log" + paramCode + ";"
+		commandCode = "/bin/bash /home/work/run_train_for_openi.sh /home/work/openi.py /tmp/log/train.log" + paramCode + ";"
 	} else if processorType == grampus.ProcessorTypeGPU {
 		if pretrainModelFileName != "" {
 			paramCode += " --ckpt_url" + "=" + workDir + "pretrainmodel/" + pretrainModelFileName