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

Merge pull request 'updated imageinference logics' (#290) from tzwang/pcm-coordinator:master into master 

Former-commit-id: eebb983c4b6290a88aa508f387cbebfb7f2ed25f
This commit is contained in:
tzwang 2024-08-28 17:25:24 +08:00
parent b9ed8c5fa7 739948d184
commit cfd0a6e168
7 changed files with 125 additions and 76 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
desc/inference/inference.api
View File

@ -3,6 +3,8 @@ syntax = "v1"
type (
/******************image inference*************************/
DeployInstance {
Id string `json:"id"`
DeployTaskId string `json:"deployTaskId"`
InstanceId string `json:"instanceId"`
InstanceName string `json:"instanceName"`
AdapterId string `json:"adapterId"`
@ -13,6 +15,9 @@ type (
ModelType string `json:"modelType"`
InferCard string `json:"inferCard"`
Status string `json:"status"`
CreateTime string `json:"createTime"`
UpdateTime string `json:"updateTime"`
ClusterType string `json:"clusterType"`
}
/******************image inference*************************/
@ -30,13 +35,13 @@ type (
/******************image inference*************************/
ImageInferenceReq {
TaskName string `json:"taskName"`
TaskDesc string `json:"taskDesc"`
ModelType string `json:"modelType"`
Instances []DeployInstance `json:"instances"`
Strategy string `json:"strategy,,optional"`
StaticWeightMap map[string]int32 `json:"staticWeightMap,optional"`
Replica int32 `json:"replicas,optional"`
TaskName string `form:"taskName"`
TaskDesc string `form:"taskDesc"`
ModelType string `form:"modelType"`
InstanceIds []int64 `form:"instanceIds"`
Strategy string `form:"strategy,,optional"`
StaticWeightMap map[string]map[string]int32 `form:"staticWeightMap,optional"`
Replica int32 `form:"replicas,optional"`
}
ImageInferenceResp {

127
internal/logic/inference/imageinferencelogic.go
View File

@ -10,7 +10,9 @@ import (
"gitlink.org.cn/JointCloud/pcm-coordinator/internal/scheduler/strategy"
"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/models"
"net/http"
"strconv"
)
type ImageInferenceLogic struct {
@ -34,21 +36,24 @@ func NewImageInferenceLogic(ctx context.Context, svcCtx *svc.ServiceContext) *Im
func (l *ImageInferenceLogic) ImageInfer(r *http.Request, req *types.ImageInferenceReq) (resp *types.ImageInferenceResp, err error) {
resp = &types.ImageInferenceResp{}
if len(req.Instances) == 0 {
if len(req.InstanceIds) == 0 {
return nil, errors.New("instances are empty")
}
opt := &option.InferOption{
TaskName: req.TaskName,
TaskDesc: req.TaskDesc,
//AdapterId: req.AdapterId,
//AiClusterIds: req.AiClusterIds,
//ModelName: req.ModelName,
ModelType: req.ModelType,
Strategy: req.Strategy,
StaticWeightMap: req.StaticWeightMap,
var instanceList []*models.AiInferDeployInstance
for _, id := range req.InstanceIds {
instance, err := l.svcCtx.Scheduler.AiStorages.GetInferDeployInstanceById(id)
if err != nil {
return nil, err
}
instanceList = append(instanceList, instance)
}
if len(instanceList) == 0 {
return nil, errors.New("instances are empty")
}
// process uploaded images
var ts []*imageInference.ImageFile
uploadedFiles := r.MultipartForm.File
@ -76,54 +81,66 @@ func (l *ImageInferenceLogic) ImageInfer(r *http.Request, req *types.ImageInfere
ts = append(ts, &t)
}
//_, ok := l.svcCtx.Scheduler.AiService.AiCollectorAdapterMap[opt.AdapterId]
//if !ok {
// return nil, errors.New("AdapterId does not exist")
//}
//
var cs []*strategy.AssignedCluster
var adapterName string
if opt.Strategy != "" {
var strat strategy.Strategy
switch opt.Strategy {
case strategy.STATIC_WEIGHT:
strat = strategy.NewStaticWeightStrategy(opt.StaticWeightMap, int32(len(ts)))
if err != nil {
return nil, err
}
default:
return nil, errors.New("no strategy has been chosen")
}
clusters, err := strat.Schedule()
if err != nil {
return nil, err
}
if clusters == nil || len(clusters) == 0 {
return nil, errors.New("clusters is nil")
}
for i := len(clusters) - 1; i >= 0; i-- {
if clusters[i].Replicas == 0 {
clusters = append(clusters[:i], clusters[i+1:]...)
}
}
name, err := l.svcCtx.Scheduler.AiStorages.GetAdapterNameById(opt.AdapterId)
if err != nil {
return nil, err
}
adapterName = name
//single adapter logic
if len(req.StaticWeightMap) != 1 {
return nil, errors.New("staticWeightMap != 1")
}
//else {
// for i, instance := range req.Instances {
//
// }
//}
adapterId := strconv.FormatInt(instanceList[0].AdapterId, 10)
staticWeightMap, ok := req.StaticWeightMap[adapterId]
if !ok {
return nil, errors.New("set staticWeightMap failed")
}
imageInfer, err := imageInference.New(imageInference.NewImageClassification(), ts, cs, req.Instances, opt, l.svcCtx.Scheduler.AiStorages, l.svcCtx.Scheduler.AiService.InferenceAdapterMap, adapterName)
// create InferOption
opt := &option.InferOption{
TaskName: req.TaskName,
TaskDesc: req.TaskDesc,
AdapterId: adapterId,
//AiClusterIds: req.AiClusterIds,
//ModelName: req.ModelName,
ModelType: req.ModelType,
Strategy: req.Strategy,
StaticWeightMap: staticWeightMap,
}
adapterName, err := l.svcCtx.Scheduler.AiStorages.GetAdapterNameById(opt.AdapterId)
if err != nil {
return nil, err
}
// set strategy
if opt.Strategy != "" {
return nil, errors.New("strategy is empty")
}
var strat strategy.Strategy
switch opt.Strategy {
case strategy.STATIC_WEIGHT:
strat = strategy.NewStaticWeightStrategy(opt.StaticWeightMap, int32(len(ts)))
if err != nil {
return nil, err
}
default:
return nil, errors.New("no strategy has been chosen")
}
clusters, err := strat.Schedule()
if err != nil {
return nil, err
}
if clusters == nil || len(clusters) == 0 {
return nil, errors.New("clusters is nil")
}
for i := len(clusters) - 1; i >= 0; i-- {
if clusters[i].Replicas == 0 {
clusters = append(clusters[:i], clusters[i+1:]...)
}
}
// create inference struct
imageInfer, err := imageInference.New(imageInference.NewImageClassification(), ts, clusters, instanceList, opt, l.svcCtx.Scheduler.AiStorages, l.svcCtx.Scheduler.AiService.InferenceAdapterMap, adapterName)
if err != nil {
return nil, err
}

27
internal/scheduler/service/inference/imageInference/imageInference.go
View File

@ -46,7 +46,7 @@ type ImageInference struct {
inference IImageInference
files []*ImageFile
clusters []*strategy.AssignedCluster
instances []types.DeployInstance
instances []*models.AiInferDeployInstance
opt *option.InferOption
storage *database.AiStorage
inferAdapter map[string]map[string]inference.ICluster
@ -58,7 +58,7 @@ func New(
inference IImageInference,
files []*ImageFile,
clusters []*strategy.AssignedCluster,
instances []types.DeployInstance,
instances []*models.AiInferDeployInstance,
opt *option.InferOption,
storage *database.AiStorage,
inferAdapter map[string]map[string]inference.ICluster,
@ -131,7 +131,7 @@ func (i *ImageInference) saveTask() (int64, error) {
return 0, err
}
i.storage.AddNoticeInfo(i.opt.AdapterId, i.adapterName, "", "", i.opt.TaskName, "create", "任务创建中")
i.storage.AddNoticeInfo("", "", "", "", i.opt.TaskName, "create", "任务创建中")
return id, nil
}
@ -197,21 +197,36 @@ func (i *ImageInference) filterClusters() ([]*FilteredCluster, error) {
var cs []*FilteredCluster
for _, cluster := range i.clusters {
var inferurls []*inference.InferUrl
var clustertype string
for _, instance := range i.instances {
if cluster.ClusterId == instance.ClusterId {
clusterId := strconv.FormatInt(instance.ClusterId, 10)
adapterId := strconv.FormatInt(instance.AdapterId, 10)
if cluster.ClusterId == clusterId {
r := http.Request{}
deployInstance, err := i.inferAdapter[instance.AdapterId][instance.ClusterId].GetInferDeployInstance(r.Context(), instance.InstanceId)
deployInstance, err := i.inferAdapter[adapterId][clusterId].GetInferDeployInstance(r.Context(), instance.InstanceId)
if err != nil {
return nil, err
continue
}
var url inference.InferUrl
url.Url = deployInstance.InferUrl
url.Card = deployInstance.InferCard
inferurls = append(inferurls, &url)
clustertype = deployInstance.ClusterType
}
}
if len(inferurls) == 0 {
continue
}
i.inference.AppendRoute(inferurls)
var f FilteredCluster
f.urls = inferurls
f.clusterName = cluster.ClusterName
f.clusterType = clustertype
f.imageNum = cluster.Replicas
cs = append(cs, &f)
}
return cs, nil

2
internal/scheduler/service/inference/textInference/textToImage.go
View File

@ -9,7 +9,7 @@ import (
)
const (
TEXTTOIMAGE = "text-to-image"
TEXTTOIMAGE = "generate_image"
TEXTTOIMAGE_AiTYPE = "14"
)

2
internal/scheduler/service/inference/textInference/textToText.go
View File

@ -79,9 +79,11 @@ func filterClusters(opt *option.InferOption, storage *database.AiStorage, inferA
wg.Done()
return
}
for i, _ := range clusterInferUrl.InferUrls {
clusterInferUrl.InferUrls[i].Url = clusterInferUrl.InferUrls[i].Url + inference.FORWARD_SLASH + CHAT
}
clusterName, _ := storage.GetClusterNameById(cId)
var f FilteredCluster

5
internal/storeLink/octopus.go
View File

@ -1154,11 +1154,16 @@ func (o *OctopusLink) GetInferDeployInstance(ctx context.Context, id string) (*i
if resp.Payload == nil {
return nil, errors.New("instance does not exist")
}
url := strings.Replace(resp.Payload.Notebook.Tasks[0].Url, FORWARD_SLASH, "", -1)
inferUrl := DOMAIN + url
ins.InstanceName = resp.Payload.Notebook.Name
ins.InstanceId = resp.Payload.Notebook.Id
ins.ClusterName = o.platform
ins.Status = resp.Payload.Notebook.Status
ins.ClusterType = TYPE_OCTOPUS
ins.InferUrl = inferUrl
return ins, nil
}

19
internal/types/types.go
View File

@ -5905,6 +5905,8 @@ type Category struct {
}
type DeployInstance struct {
Id string `json:"id"`
DeployTaskId string `json:"deployTaskId"`
InstanceId string `json:"instanceId"`
InstanceName string `json:"instanceName"`
AdapterId string `json:"adapterId"`
@ -5915,6 +5917,9 @@ type DeployInstance struct {
ModelType string `json:"modelType"`
InferCard string `json:"inferCard"`
Status string `json:"status"`
CreateTime string `json:"createTime"`
UpdateTime string `json:"updateTime"`
ClusterType string `json:"clusterType"`
}
type ModelTypesResp struct {
@ -5930,13 +5935,13 @@ type ModelNamesResp struct {
}
type ImageInferenceReq struct {
TaskName string `json:"taskName"`
TaskDesc string `json:"taskDesc"`
ModelType string `json:"modelType"`
Instances []DeployInstance `json:"instances"`
Strategy string `json:"strategy,,optional"`
StaticWeightMap map[string]int32 `json:"staticWeightMap,optional"`
Replica int32 `json:"replicas,optional"`
TaskName string `form:"taskName"`
TaskDesc string `form:"taskDesc"`
ModelType string `form:"modelType"`
InstanceIds []int64 `form:"instanceIds"`
Strategy string `form:"strategy,,optional"`
StaticWeightMap map[string]map[string]int32 `form:"staticWeightMap,optional"`
Replica int32 `form:"replicas,optional"`
}
type ImageInferenceResp struct {