floraachy
/
pcm-coordinator
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

modified octopus, shuguangai implications 

Former-commit-id: 62cac010a042f1c1fa932fad479eccfcb6c58156
This commit is contained in:
tzwang 2024-02-28 16:38:03 +08:00
parent e2ffaeee37
commit 1356ad5932
3 changed files with 26 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
api/internal/storeLink/octopus.go
View File

@ -50,6 +50,8 @@ const (
EnflameT20 = 128
BASE_TOPS = 128
CAMBRICON = "cambricon"
TRAIN_CMD = "cd /code; python train.py"
VERSION = "V1"
)
var (
@ -163,6 +165,10 @@ func (o *OctopusLink) SubmitTask(imageId string, cmd string, envs []string, para
Envs: envMap,
},
},
DataSetId: datasetsId,
DataSetVersion: VERSION,
AlgorithmId: algorithmId,
AlgorithmVersion: VERSION,
},
}
resp, err := o.svcCtx.OctopusRpc.CreateTrainJob(o.ctx, req)
@ -530,6 +536,9 @@ func (o *OctopusLink) generateAlgorithmId(option *option.AiOption) error {
}
func (o *OctopusLink) generateCmd(option *option.AiOption) error {
if option.Cmd == "" {
option.Cmd = TRAIN_CMD
}
return nil
}

4
api/internal/storeLink/shuguangHpc.go
View File

@ -192,7 +192,7 @@ func (s ShuguangHpc) SubmitTask(imageId string, cmd string, envs []string, param
},
}
updateRequestByResourceId(resourceId, req)
updateSGHpcRequestByResourceId(resourceId, req)
resp, err := s.svcCtx.ACRpc.SubmitJob(s.ctx, req)
if err != nil {
@ -261,7 +261,7 @@ func (s ShuguangHpc) DeleteTask(taskId string) (interface{}, error) {
return resp, nil
}
func updateRequestByResourceId(resourceId string, req *hpcAC.SubmitJobReq) {
func updateSGHpcRequestByResourceId(resourceId string, req *hpcAC.SubmitJobReq) {
spec := RESOURCEMAP[resourceId]
req.MapAppJobInfo.GAP_NNODE = spec.GAP_NNODE
req.MapAppJobInfo.GAP_NPROC = spec.GAP_NPROC

23
api/internal/storeLink/shuguangai.go
View File

@ -28,8 +28,7 @@ import (
)
const (
RAM_SIZE_1G = 1024 // 1G
WORKER_RAM_SIZE = 10240 // 10G
RAM_SIZE_1G = 1024 // 1G
WORKER_NUMBER = 1
WORKER_CPU_NUMBER = 5
WORKER_GPU_NUMBER = 1
@ -46,7 +45,7 @@ const (
DATASETS_DIR = "/work/home/acgnnmfbwo/pcmv1/dataset"
)
var RESOURCESGMAP = map[string]ResourceSpecSG{
var RESOURCESGAIMAP = map[string]ResourceSpecSGAI{
"WodTB2rJ8SobMgQ1nrtR245jxOrsovFi": {
CPU: 1,
GPU: 1,
@ -82,7 +81,7 @@ var RESOURCESPECSAI = map[string]string{
"jeYBVPwyIALjVYNzHvysh2o5CsBpBLp2": "CPU:5, DCU:2, RAM:10G",
}
type ResourceSpecSG struct {
type ResourceSpecSGAI struct {
CPU int64
GPU int64
RAM int64
@ -123,7 +122,8 @@ func (s *ShuguangAi) QueryImageList() (interface{}, error) {
func (s *ShuguangAi) SubmitPytorchTask(imageId string, cmd string, envs []string, params []string, resourceId string) (interface{}, error) {
//判断是否resourceId匹配自定义资源Id
if resourceId != SHUGUANGAI_CUSTOM_RESOURCE_ID {
_, isMapContainsKey := RESOURCESPECSAI[resourceId]
if !isMapContainsKey {
return nil, errors.New("shuguangAi资源Id不存在")
}
@ -159,15 +159,15 @@ func (s *ShuguangAi) SubmitPytorchTask(imageId string, cmd string, envs []string
Version: imageResp.Image.Version,
ImagePath: imageResp.Image.Path,
WorkerNumber: WORKER_NUMBER,
WorkerCpuNumber: WORKER_CPU_NUMBER,
WorkerGpuNumber: WORKER_GPU_NUMBER,
WorkerRamSize: WORKER_RAM_SIZE,
ResourceGroup: RESOURCE_GROUP,
TimeoutLimit: TimeoutLimit,
PythonCodePath: PythonCodePath,
PythonArg: pythonArg,
},
}
updateSGAIRequestByResourceId(resourceId, req)
resp, err := s.svcCtx.ACRpc.SubmitPytorchTask(s.ctx, req)
if err != nil {
return nil, err
@ -176,6 +176,13 @@ func (s *ShuguangAi) SubmitPytorchTask(imageId string, cmd string, envs []string
return resp, nil
}
func updateSGAIRequestByResourceId(resourceId string, req *hpcAC.SubmitPytorchTaskReq) {
spec := RESOURCESGAIMAP[resourceId]
req.Params.WorkerCpuNumber = spec.CPU
req.Params.WorkerGpuNumber = spec.GPU
req.Params.WorkerRamSize = spec.RAM
}
func (s *ShuguangAi) SubmitTensorflowTask(imageId string, cmd string, envs []string, params []string, resourceId string) (interface{}, error) {
//req := &hpcAC.SubmitTensorflowTaskReq{
// Params: &hpcAC.SubmitTensorflowTaskParams{