From 99b2c851255c8bad881300517a8feeb59cf072f1 Mon Sep 17 00:00:00 2001
From: lewis <747342561@qq.com>
Date: Tue, 31 May 2022 18:02:30 +0800
Subject: [PATCH] view
---
models/cloudbrain.go | 14 +++++---
modules/grampus/grampus.go | 3 +-
modules/grampus/resty.go | 2 +-
options/locale/locale_en-US.ini | 5 +++
options/locale/locale_zh-CN.ini | 5 +++
routers/api/v1/api.go | 2 +-
routers/repo/cloudbrain.go | 3 ++
routers/repo/grampus.go | 38 ++++++++++----------
routers/routes/routes.go | 2 +-
templates/repo/cloudbrain/trainjob/new.tmpl | 13 +++++++
templates/repo/grampus/trainjob/gpu/new.tmpl | 13 +++++++
templates/repo/grampus/trainjob/npu/new.tmpl | 13 +++++++
templates/repo/grampus/trainjob/show.tmpl | 13 ++++++-
templates/repo/modelarts/trainjob/new.tmpl | 13 +++++++
14 files changed, 110 insertions(+), 29 deletions(-)
diff --git a/models/cloudbrain.go b/models/cloudbrain.go
index 06fbea5b3..97fa69e0d 100755
--- a/models/cloudbrain.go
+++ b/models/cloudbrain.go
@@ -147,6 +147,7 @@ type Cloudbrain struct {
ComputeResource string //计算资源，例如npu
EngineID int64 //引擎id
ImageID string //grampus image_id
+ AiCenter string //grampus ai center: center_id+center_name
TrainUrl string //输出模型的obs路径
BranchName string //分支名称
@@ -1224,11 +1225,14 @@ type GrampusStopJobResponse struct {
}
type GrampusTasks struct {
- Command string `json:"command"`
- Name string `json:"name"`
- ImageId string `json:"imageId"`
- ResourceSpecId string `json:"resourceSpecId"`
- ImageUrl string `json:"imageUrl"`
+ Command string `json:"command"`
+ Name string `json:"name"`
+ ImageId string `json:"imageId"`
+ ResourceSpecId string `json:"resourceSpecId"`
+ ImageUrl string `json:"imageUrl"`
+ CenterID []string `json:"centerID"`
+ CenterName []string `json:"centerName"`
+ ReplicaNum int `json:"replicaNum"`
}
type CreateGrampusJobRequest struct {
diff --git a/modules/grampus/grampus.go b/modules/grampus/grampus.go
index 95985e533..a0f398115 100755
--- a/modules/grampus/grampus.go
+++ b/modules/grampus/grampus.go
@@ -40,7 +40,7 @@ const (
ProcessorTypeNPU = "npu.huawei.com/NPU"
ProcessorTypeGPU = "nvidia.com/gpu"
- CommandPrepareScript = "pwd;cd /tmp;mkdir output;mkdir code;mkdir dataset;wget -q https://git.openi.org.cn/lewis/script_for_grampus/archive/master.zip;unzip -q master.zip;cd script_for_grampus;chmod 777 sync_for_arm uploader_for_grampus;"
+ CommandPrepareScript = "cd /tmp;mkdir -p output;mkdir -p code;mkdir -p dataset;wget -q https://git.openi.org.cn/lewis/script_for_grampus/archive/master.zip;unzip -q master.zip;cd script_for_grampus;chmod 777 sync_for_arm uploader_for_grampus;"
CodeArchiveName = "master.zip"
)
@@ -92,6 +92,7 @@ func GenerateTrainJob(ctx *context.Context, req *GenerateTrainJobReq) (err error
ResourceSpecId: req.ResourceSpecId,
ImageId: req.ImageId,
ImageUrl: req.ImageUrl,
+ ReplicaNum: 0,
},
},
})
diff --git a/modules/grampus/resty.go b/modules/grampus/resty.go
index 183afb853..bd64ace8f 100755
--- a/modules/grampus/resty.go
+++ b/modules/grampus/resty.go
@@ -228,7 +228,7 @@ func GetTrainJobLog(jobID string) (string, error) {
res, err := client.R().
SetAuthToken(TOKEN).
SetResult(&logContent).
- Get(HOST + urlTrainJob + "/" + jobID + "/log")
+ Get(HOST + urlTrainJob + "/" + jobID + "/task/0/replica/0/log")
if err != nil {
return logContent, fmt.Errorf("resty GetTrainJobLog: %v", err)
diff --git a/options/locale/locale_en-US.ini b/options/locale/locale_en-US.ini
index c52a369ce..3166dafb7 100755
--- a/options/locale/locale_en-US.ini
+++ b/options/locale/locale_en-US.ini
@@ -1170,6 +1170,8 @@ model.manage.sava_model = Sava Model
model.manage.model_manage = ModelManage
model.manage.model_accuracy = Model Accuracy
+grampus.train_job.ai_center = AI Center
+
template.items = Template Items
template.git_content = Git Content (Default Branch)
template.git_hooks = Git Hooks
@@ -3013,6 +3015,9 @@ Platform_Tutorial = Tutorial
foot.advice_feedback = Feedback
[cloudbrain]
+resource_cluster = Resource Cluster
+resource_cluster_openi = OpenI Resource Cluster
+resource_cluster_c2net = China Computing NET
compute_resource = Computing resources
task_name = Task name
task_type = Task type
diff --git a/options/locale/locale_zh-CN.ini b/options/locale/locale_zh-CN.ini
index cb1c7565a..e9b6a5280 100755
--- a/options/locale/locale_zh-CN.ini
+++ b/options/locale/locale_zh-CN.ini
@@ -1180,6 +1180,8 @@ model.manage.sava_model = 保存模型
model.manage.model_manage = 模型管理
model.manage.model_accuracy = 模型精度
+grampus.train_job.ai_center=ai计算中心
+
template.items=模板选项
template.git_content=Git数据(默认分支)
template.git_hooks=Git 钩子
@@ -3023,6 +3025,9 @@ Platform_Tutorial=新手指引
foot.advice_feedback = 意见反馈
[cloudbrain]
+resource_cluster = 算力集群
+resource_cluster_openi = 启智集群
+resource_cluster_c2net = 智算集群
compute_resource = 计算资源
task_name = 任务名称
task_type = 任务类型
diff --git a/routers/api/v1/api.go b/routers/api/v1/api.go
index 471f8be7e..f6153e811 100755
--- a/routers/api/v1/api.go
+++ b/routers/api/v1/api.go
@@ -935,9 +935,9 @@ func RegisterRoutes(m *macaron.Macaron) {
})
}, reqRepoReader(models.UnitTypeCloudBrain))
m.Group("/grampus", func() {
- m.Get("/:id", repo.GetCloudbrainTask)
m.Group("/train-job", func() {
m.Group("/:jobid", func() {
+ m.Get("", repo.GetModelArtsTrainJobVersion)
m.Post("/stop_version", cloudbrain.AdminOrOwnerOrJobCreaterRightForTrain, repo_ext.GrampusStopJob)
m.Get("/log", repo_ext.GrampusGetLog)
})
diff --git a/routers/repo/cloudbrain.go b/routers/repo/cloudbrain.go
index 69cd24901..525dd07bb 100755
--- a/routers/repo/cloudbrain.go
+++ b/routers/repo/cloudbrain.go
@@ -1477,6 +1477,9 @@ func SyncCloudbrainStatus() {
}
if result != nil {
+ if len(result.JobInfo.Tasks[0].CenterID) == 1 && len(result.JobInfo.Tasks[0].CenterName) == 1 {
+ task.AiCenter = result.JobInfo.Tasks[0].CenterID[0] + "+" + result.JobInfo.Tasks[0].CenterName[0]
+ }
task.Status = grampus.TransTrainJobStatus(result.JobInfo.Status)
task.Duration = result.JobInfo.RunSec
task.TrainJobDuration = models.ConvertDurationToStr(task.Duration)
diff --git a/routers/repo/grampus.go b/routers/repo/grampus.go
index 542b1c386..d25a75c68 100755
--- a/routers/repo/grampus.go
+++ b/routers/repo/grampus.go
@@ -9,7 +9,6 @@ import (
"code.gitea.io/gitea/modules/util"
"encoding/json"
"errors"
- "fmt"
"github.com/unknwon/com"
"io/ioutil"
"net/http"
@@ -458,11 +457,14 @@ func GrampusTrainJobShow(ctx *context.Context) {
result, err := grampus.GetJob(task.JobID)
if err != nil {
log.Error("GetJob failed:" + err.Error())
- ctx.ServerError("GetJob failed", err)
- return
+ //ctx.ServerError("GetJob failed", err)
+ //return
}
if result != nil {
+ if len(result.JobInfo.Tasks[0].CenterID) == 1 && len(result.JobInfo.Tasks[0].CenterName) == 1 {
+ task.AiCenter = result.JobInfo.Tasks[0].CenterID[0] + "+" + result.JobInfo.Tasks[0].CenterName[0]
+ }
task.Status = grampus.TransTrainJobStatus(result.JobInfo.Status)
if task.Status != result.JobInfo.Status || result.JobInfo.Status == models.GrampusStatusRunning {
task.Duration = result.JobInfo.RunSec
@@ -508,6 +510,13 @@ func GrampusTrainJobShow(ctx *context.Context) {
taskList = append(taskList, task)
ctx.Data["version_list_task"] = taskList
+ ctx.Data["canDownload"] = cloudbrain.CanModifyJob(ctx, task)
+
+ aiCenterInfo := strings.Split(task.AiCenter, "+")
+ if len(aiCenterInfo) == 2 {
+ ctx.Data["ai_center"] = aiCenterInfo[1]
+ }
+
ctx.HTML(http.StatusOK, tplGrampusTrainJobShow)
}
@@ -553,7 +562,6 @@ func generateCommand(repoName, processorType, codeObsPath, dataObsPath, bootFile
toolUnzip = "tar -zxvf "
}
commandUnzip := "cd /tmp/dataset;" + toolUnzip + datasetName + ";cd /tmp/code;unzip -q master.zip;"
- commandUnzip += "cd /tmp/dataset/" + strings.TrimSuffix(datasetName, ".zip") + ";ls;"
command += commandUnzip
//exec code
@@ -579,25 +587,17 @@ func generateCommand(repoName, processorType, codeObsPath, dataObsPath, bootFile
commandCode := "cd /tmp/code/" + repoName + ";python " + bootFile + paramCode + ";"
command += commandCode
+ //get exec result
+ commandGetRes := "result=$?;"
+ command += commandGetRes
+
//upload models
commandUpload := "cd /tmp/script_for_grampus/;./uploader_for_grampus " + setting.Bucket + " " + outputObsPath + " " + "/tmp/output/;"
command += commandUpload
- return command, nil
-}
-
-func generateCommandObsDownloadFile(srcObsFile, dstLocalDir string) (string, error) {
- var command string
-
- command = "python;"
- command += "from modelarts.session import Session \n"
- command += fmt.Sprintf("session = Session(access_key='%s',secret_key='%s', project_id='%s', region_name='%s') \n", setting.AccessKeyID, setting.SecretAccessKey, setting.ProjectID, setting.Location)
-
- if util.IsDir(srcObsFile) {
- command += fmt.Sprintf("session.obs.download_dir(src_obs_dir=\"%s\", dst_local_dir=\"%s\") \n", srcObsFile, dstLocalDir)
- } else {
- command += fmt.Sprintf("session.obs.download_file(src_obs_file=\"%s\", dst_local_dir=\"%s\") \n", srcObsFile, dstLocalDir)
- }
+ //check exec result
+ commandCheckRes := " [[ result -eq 0 ]] && echo success || ls failed;"
+ command += commandCheckRes
return command, nil
}
diff --git a/routers/routes/routes.go b/routers/routes/routes.go
index 64566a1d7..ab9f12205 100755
--- a/routers/routes/routes.go
+++ b/routers/routes/routes.go
@@ -1089,7 +1089,7 @@ func RegisterRoutes(m *macaron.Macaron) {
m.Get("", reqRepoCloudBrainReader, repo.GrampusTrainJobShow)
m.Post("/stop", cloudbrain.AdminOrOwnerOrJobCreaterRight, repo.GrampusStopJob)
m.Post("/del", cloudbrain.AdminOrOwnerOrJobCreaterRightForTrain, repo.GrampusTrainJobDel)
- m.Get("/download_model", cloudbrain.AdminOrOwnerOrJobCreaterRightForTrain, repo.CloudBrainDownloadModel)
+ m.Get("/model_download", cloudbrain.AdminOrJobCreaterRightForTrain, repo.ModelDownload)
})
m.Group("/gpu", func() {
m.Get("/create", reqWechatBind, reqRepoCloudBrainWriter, repo.GrampusTrainJobGPUNew)
diff --git a/templates/repo/cloudbrain/trainjob/new.tmpl b/templates/repo/cloudbrain/trainjob/new.tmpl
index 39315cfad..4eff7c21c 100755
--- a/templates/repo/cloudbrain/trainjob/new.tmpl
+++ b/templates/repo/cloudbrain/trainjob/new.tmpl
@@ -82,6 +82,19 @@
+
+
+
+