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

modified ai option 

Former-commit-id: dc33df489a557efeecabb26e01f422b97311212a
This commit is contained in:
tzwang 2024-03-01 17:07:41 +08:00
parent bfe7597644
commit 7487707f3d
3 changed files with 118 additions and 90 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
api/internal/scheduler/schedulers/option/aiOption.go
View File

@ -2,6 +2,7 @@ package option
type AiOption struct {
AiClusterId string // shuguangAi /octopus ClusterId
TaskName string
ResourceType string // cpu/gpu/compute card
TaskType string // pytorch/tensorflow/mindspore
DatasetsName string // mnist/imageNet/iris
@ -10,6 +11,7 @@ type AiOption struct {
Tops float64
ComputeCard string
CodeType string
AlgorithmName string
ImageId string
SpecId string
@ -23,6 +25,7 @@ type AiOption struct {
Params []string
Datasets string
Code string
AlgorithmCode string
Image string
Model interface{}
}

170
api/internal/storeLink/octopus.go
View File

@ -351,91 +351,7 @@ func (o *OctopusLink) generateResourceId(option *option.AiOption) error {
}
}
return nil
}
func setResourceIdByCard(option *option.AiOption, specs *octopus.GetResourceSpecsResp, computeCard string) error {
if option.Tops == 0 {
for _, spec := range specs.TrainResourceSpecs {
if spec.Price == 1 {
ns := strings.Split(spec.Name, COMMA)
cardSpecs := strings.Split(ns[0], STAR)
if cardSpecs[1] == computeCard {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
} else {
continue
}
}
} else {
cardNum := math.Ceil(option.Tops / float64(BASE_TOPS))
for _, spec := range specs.TrainResourceSpecs {
if option.Tops < BASE_TOPS {
if spec.Price == 1 {
ns := strings.Split(spec.Name, COMMA)
cardSpecs := strings.Split(ns[0], STAR)
if cardSpecs[1] == computeCard {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
} else {
continue
}
} else {
ns := strings.Split(spec.Name, COMMA)
if len(ns) != 4 {
continue
}
cardSpecs := strings.Split(ns[0], STAR)
if cardSpecs[1] != computeCard {
continue
}
s, err := strconv.ParseFloat(cardSpecs[0], 64)
if err != nil {
return err
}
switch computeCard {
case GCU:
if cardNum == s { // 1, 4, 8
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
if 1 < cardNum && cardNum <= 4 && s == 4 {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
if 4 < cardNum && s == 8 {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
case MLU: // 1, 2, 4
if cardNum/2 == s {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
if 1 < cardNum/2 && cardNum/2 <= 2 && s == 2 {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
if 2 < cardNum/2 && s == 4 {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
}
}
}
}
return errors.New("set ResourceId error")
return errors.New("failed to get ResourceId")
}
func (o *OctopusLink) generateDatasetsId(option *option.AiOption) error {
@ -552,3 +468,87 @@ func (o *OctopusLink) generateParams(option *option.AiOption) error {
return nil
}
func setResourceIdByCard(option *option.AiOption, specs *octopus.GetResourceSpecsResp, computeCard string) error {
if option.Tops == 0 {
for _, spec := range specs.TrainResourceSpecs {
if spec.Price == 1 {
ns := strings.Split(spec.Name, COMMA)
cardSpecs := strings.Split(ns[0], STAR)
if cardSpecs[1] == computeCard {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
} else {
continue
}
}
} else {
cardNum := math.Ceil(option.Tops / float64(BASE_TOPS))
for _, spec := range specs.TrainResourceSpecs {
if option.Tops < BASE_TOPS {
if spec.Price == 1 {
ns := strings.Split(spec.Name, COMMA)
cardSpecs := strings.Split(ns[0], STAR)
if cardSpecs[1] == computeCard {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
} else {
continue
}
} else {
ns := strings.Split(spec.Name, COMMA)
if len(ns) != 4 {
continue
}
cardSpecs := strings.Split(ns[0], STAR)
if cardSpecs[1] != computeCard {
continue
}
s, err := strconv.ParseFloat(cardSpecs[0], 64)
if err != nil {
return err
}
switch computeCard {
case GCU:
if cardNum == s { // 1, 4, 8
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
if 1 < cardNum && cardNum <= 4 && s == 4 {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
if 4 < cardNum && s == 8 {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
case MLU: // 1, 2, 4
if cardNum/2 == s {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
if 1 < cardNum/2 && cardNum/2 <= 2 && s == 2 {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
if 2 < cardNum/2 && s == 4 {
option.ResourceId = spec.Id
option.ComputeCard = computeCard
return nil
}
}
}
}
}
return errors.New("set ResourceId error")
}

29
api/internal/storeLink/shuguangai.go
View File

@ -31,6 +31,7 @@ const (
RAM_SIZE_1G = 1024 // 1G
WORKER_NUMBER = 1
DCU = "dcu"
DCU_TOPS = 24.5
PYTORCH = "Pytorch"
TASK_PYTORCH_PREFIX = "PytorchTask"
TENSORFLOW = "Tensorflow"
@ -345,8 +346,27 @@ func (s *ShuguangAi) GenerateSubmitParams(option *option.AiOption) error {
}
func (s *ShuguangAi) generateResourceId(option *option.AiOption) error {
if option.ResourceType == "" {
return errors.New("ResourceType not set")
}
return nil
if option.ResourceType == CPU {
option.ResourceId = "WodTB2rJ8SobMgQ1nrtR245jxOrsovFi"
}
if option.ResourceType == CARD {
if option.Tops == 0 {
option.ResourceId = "WodTB2rJ8SobMgQ1nrtR245jxOrsovFi"
}
if option.Tops > DCU_TOPS {
option.ResourceId = "jeYBVPwyIALjVYNzHvysh2o5CsBpBLp2"
}
//Todo add more dcu specs
}
return errors.New("failed to get ResourceId")
}
func (s *ShuguangAi) generateImageId(option *option.AiOption) error {
@ -365,7 +385,12 @@ func (s *ShuguangAi) generateImageId(option *option.AiOption) error {
if resp.Code != "0" {
return errors.New("failed to get imageId")
}
return nil
if option.ResourceType == CPU {
}
return errors.New("failed to get ImageId")
}
func (s *ShuguangAi) generateAlgorithmId(option *option.AiOption) error {