Merge pull request #255 from kerthcet/cleanup/update-flavor-field

InftyAI-Agent · web-flow · commit 82833b72a2e7 · 2025-01-24T23:23:24.000+08:00
Rename inferenceFlavorClaims to inferenceFlavors
diff --git a/api/core/v1alpha1/model_types.go b/api/core/v1alpha1/model_types.go
@@ -139,12 +139,12 @@ type ModelName string
 type ModelClaim struct {
 	// ModelName represents the name of the Model.
 	ModelName ModelName `json:"modelName,omitempty"`
-	// InferenceFlavorClaims represents a list of flavors with fungibility support
+	// InferenceFlavors represents a list of flavors with fungibility support
 	// to serve the model.
 	// If set, The flavor names should be a subset of the model configured flavors.
 	// If not set, Model configured flavors will be used by default.
 	// +optional
-	InferenceFlavorClaims []FlavorName `json:"inferenceFlavorClaims,omitempty"`
+	InferenceFlavors []FlavorName `json:"inferenceFlavors,omitempty"`
 }
 
 type ModelRole string
@@ -182,12 +182,12 @@ type ModelClaims struct {
 	// is draft model.
 	// +kubebuilder:validation:MinItems=1
 	Models []ModelRef `json:"models,omitempty"`
-	// InferenceFlavorClaims represents a list of flavors with fungibility supported
+	// InferenceFlavors represents a list of flavor names with fungibility supported
 	// to serve the model.
 	// - If not set, always apply with the 0-index model by default.
 	// - If set, will lookup the flavor names following the model orders.
 	// +optional
-	InferenceFlavorClaims []FlavorName `json:"inferenceFlavorClaims,omitempty"`
+	InferenceFlavors []FlavorName `json:"inferenceFlavors,omitempty"`
 }
 
 // ModelSpec defines the desired state of Model
diff --git a/api/core/v1alpha1/zz_generated.deepcopy.go b/api/core/v1alpha1/zz_generated.deepcopy.go
diff --git a/client-go/applyconfiguration/core/v1alpha1/modelclaim.go b/client-go/applyconfiguration/core/v1alpha1/modelclaim.go
diff --git a/client-go/applyconfiguration/core/v1alpha1/modelclaims.go b/client-go/applyconfiguration/core/v1alpha1/modelclaims.go
diff --git a/config/crd/bases/inference.llmaz.io_playgrounds.yaml b/config/crd/bases/inference.llmaz.io_playgrounds.yaml
@@ -880,9 +880,9 @@ spec:
                   of modelClaims. Most of the time, modelClaim is enough.
                   ModelClaim and modelClaims are exclusive configured.
                 properties:
-                  inferenceFlavorClaims:
+                  inferenceFlavors:
                     description: |-
-                      InferenceFlavorClaims represents a list of flavors with fungibility support
+                      InferenceFlavors represents a list of flavors with fungibility support
                       to serve the model.
                       If set, The flavor names should be a subset of the model configured flavors.
                       If not set, Model configured flavors will be used by default.
@@ -899,9 +899,9 @@ spec:
                   use cases like speculative-decoding.
                   ModelClaims and modelClaim are exclusive configured.
                 properties:
-                  inferenceFlavorClaims:
+                  inferenceFlavors:
                     description: |-
-                      InferenceFlavorClaims represents a list of flavors with fungibility supported
+                      InferenceFlavors represents a list of flavor names with fungibility supported
                       to serve the model.
                       - If not set, always apply with the 0-index model by default.
                       - If set, will lookup the flavor names following the model orders.
diff --git a/config/crd/bases/inference.llmaz.io_services.yaml b/config/crd/bases/inference.llmaz.io_services.yaml
@@ -48,9 +48,9 @@ spec:
                 description: ModelClaims represents multiple claims for different
                   models.
                 properties:
-                  inferenceFlavorClaims:
+                  inferenceFlavors:
                     description: |-
-                      InferenceFlavorClaims represents a list of flavors with fungibility supported
+                      InferenceFlavors represents a list of flavor names with fungibility supported
                       to serve the model.
                       - If not set, always apply with the 0-index model by default.
                       - If set, will lookup the flavor names following the model orders.
diff --git a/docs/examples/multi-nodes/playground.yaml b/docs/examples/multi-nodes/playground.yaml
@@ -6,7 +6,7 @@ spec:
   replicas: 1
   modelClaim:
     modelName: llama3-405b-instruct
-    inferenceFlavorClaims:
+    inferenceFlavors:
       - a100-80gb # actually no need to specify this since we have only one flavor
   backendRuntimeConfig:
     resources:
diff --git a/pkg/controller/inference/playground_controller.go b/pkg/controller/inference/playground_controller.go
@@ -207,7 +207,7 @@ func buildServiceApplyConfiguration(models []*coreapi.OpenModel, playground *inf
 	if playground.Spec.ModelClaim != nil {
 		claim = coreclientgo.ModelClaims().
 			WithModels(coreclientgo.ModelRef().WithName(playground.Spec.ModelClaim.ModelName).WithRole(coreapi.MainRole)).
-			WithInferenceFlavorClaims(playground.Spec.ModelClaim.InferenceFlavorClaims...)
+			WithInferenceFlavors(playground.Spec.ModelClaim.InferenceFlavors...)
 	} else {
 		mrs := []*coreclientgo.ModelRefApplyConfiguration{}
 		for _, model := range playground.Spec.ModelClaims.Models {
@@ -221,7 +221,7 @@ func buildServiceApplyConfiguration(models []*coreapi.OpenModel, playground *inf
 
 		claim = coreclientgo.ModelClaims().
 			WithModels(mrs...).
-			WithInferenceFlavorClaims(playground.Spec.ModelClaims.InferenceFlavorClaims...)
+			WithInferenceFlavors(playground.Spec.ModelClaims.InferenceFlavors...)
 	}
 
 	spec.WithModelClaims(claim)
diff --git a/pkg/controller/inference/service_controller.go b/pkg/controller/inference/service_controller.go
@@ -187,9 +187,9 @@ func injectModelFlavor(template *corev1.PodTemplateSpec, model *coreapi.OpenMode
 	}
 
 	flavorName := model.Spec.InferenceConfig.Flavors[0].Name
-	if len(service.Spec.ModelClaims.InferenceFlavorClaims) > 0 {
+	if len(service.Spec.ModelClaims.InferenceFlavors) > 0 {
 		// We only support the same resource request right now, so 0-index flavor is enough.
-		flavorName = service.Spec.ModelClaims.InferenceFlavorClaims[0]
+		flavorName = service.Spec.ModelClaims.InferenceFlavors[0]
 	}
 
 	for i, flavor := range model.Spec.InferenceConfig.Flavors {
@@ -222,8 +222,8 @@ func modelLabels(model *coreapi.OpenModel) map[string]string {
 
 func modelAnnotations(service *inferenceapi.Service) map[string]string {
 	var values string
-	for i, value := range service.Spec.ModelClaims.InferenceFlavorClaims {
-		if i == len(service.Spec.ModelClaims.InferenceFlavorClaims)-1 {
+	for i, value := range service.Spec.ModelClaims.InferenceFlavors {
+		if i == len(service.Spec.ModelClaims.InferenceFlavors)-1 {
 			values += string(value)
 		} else {
 			values += string(value) + ","
diff --git a/pkg/controller_helper/helper.go b/pkg/controller_helper/helper.go
@@ -95,9 +95,9 @@ func fetchModels(ctx context.Context, k8sClient client.Client, mrs []coreapi.Mod
 func FirstAssignedFlavor(model *coreapi.OpenModel, playground *inferenceapi.Playground) []coreapi.Flavor {
 	var flavors []coreapi.FlavorName
 	if playground.Spec.ModelClaim != nil {
-		flavors = playground.Spec.ModelClaim.InferenceFlavorClaims
+		flavors = playground.Spec.ModelClaim.InferenceFlavors
 	} else {
-		flavors = playground.Spec.ModelClaims.InferenceFlavorClaims
+		flavors = playground.Spec.ModelClaims.InferenceFlavors
 	}
 
 	if len(flavors) == 0 && (model.Spec.InferenceConfig == nil || len(model.Spec.InferenceConfig.Flavors) == 0) {
diff --git a/test/util/validation/validate_playground.go b/test/util/validation/validate_playground.go
@@ -45,15 +45,15 @@ func validateModelClaim(models []*coreapi.OpenModel, playground *inferenceapi.Pl
 		if playground.Spec.ModelClaim.ModelName != service.Spec.ModelClaims.Models[0].Name {
 			return fmt.Errorf("expected modelName %s, got %s", playground.Spec.ModelClaim.ModelName, service.Spec.ModelClaims.Models[0].Name)
 		}
-		if diff := cmp.Diff(playground.Spec.ModelClaim.InferenceFlavorClaims, service.Spec.ModelClaims.InferenceFlavorClaims); diff != "" {
-			return fmt.Errorf("unexpected flavors, want %v, got %v", playground.Spec.ModelClaim.InferenceFlavorClaims, service.Spec.ModelClaims.InferenceFlavorClaims)
+		if diff := cmp.Diff(playground.Spec.ModelClaim.InferenceFlavors, service.Spec.ModelClaims.InferenceFlavors); diff != "" {
+			return fmt.Errorf("unexpected flavors, want %v, got %v", playground.Spec.ModelClaim.InferenceFlavors, service.Spec.ModelClaims.InferenceFlavors)
 		}
 	} else if playground.Spec.ModelClaims != nil {
 		if diff := cmp.Diff(*playground.Spec.ModelClaims, service.Spec.ModelClaims); diff != "" {
 			return fmt.Errorf("expected modelClaims, want %v, got %v", *playground.Spec.ModelClaims, service.Spec.ModelClaims)
 		}
-		if diff := cmp.Diff(playground.Spec.ModelClaims.InferenceFlavorClaims, service.Spec.ModelClaims.InferenceFlavorClaims); diff != "" {
-			return fmt.Errorf("unexpected flavors, want %v, got %v", playground.Spec.ModelClaim.InferenceFlavorClaims, service.Spec.ModelClaims.InferenceFlavorClaims)
+		if diff := cmp.Diff(playground.Spec.ModelClaims.InferenceFlavors, service.Spec.ModelClaims.InferenceFlavors); diff != "" {
+			return fmt.Errorf("unexpected flavors, want %v, got %v", playground.Spec.ModelClaim.InferenceFlavors, service.Spec.ModelClaims.InferenceFlavors)
 		}
 	}
 
diff --git a/test/util/validation/validate_service.go b/test/util/validation/validate_service.go
@@ -164,8 +164,8 @@ func ValidateModelLoader(model *coreapi.OpenModel, index int, template corev1.Po
 
 func ValidateModelFlavor(service *inferenceapi.Service, model *coreapi.OpenModel, workload *lws.LeaderWorkerSet) error {
 	flavorName := model.Spec.InferenceConfig.Flavors[0].Name
-	if len(service.Spec.ModelClaims.InferenceFlavorClaims) > 0 {
-		flavorName = service.Spec.ModelClaims.InferenceFlavorClaims[0]
+	if len(service.Spec.ModelClaims.InferenceFlavors) > 0 {
+		flavorName = service.Spec.ModelClaims.InferenceFlavors[0]
 	}
 
 	for _, flavor := range model.Spec.InferenceConfig.Flavors {
diff --git a/test/util/wrapper/playground.go b/test/util/wrapper/playground.go
@@ -67,7 +67,7 @@ func (w *PlaygroundWrapper) ModelClaim(modelName string, flavorNames ...string)
 	}
 
 	if len(names) > 0 {
-		w.Spec.ModelClaim.InferenceFlavorClaims = names
+		w.Spec.ModelClaim.InferenceFlavors = names
 	}
 	return w
 }
@@ -87,7 +87,7 @@ func (w *PlaygroundWrapper) ModelClaims(modelNames []string, roles []string, fla
 	}
 
 	if len(fNames) > 0 {
-		w.Spec.ModelClaims.InferenceFlavorClaims = fNames
+		w.Spec.ModelClaims.InferenceFlavors = fNames
 	}
 	return w
 }
diff --git a/test/util/wrapper/service.go b/test/util/wrapper/service.go
@@ -59,7 +59,7 @@ func (w *ServiceWrapper) ModelClaims(modelNames []string, roles []string, flavor
 	}
 
 	if len(fNames) > 0 {
-		w.Spec.ModelClaims.InferenceFlavorClaims = fNames
+		w.Spec.ModelClaims.InferenceFlavors = fNames
 	}
 	return w
 }

Original file line number	Diff line number	Diff line change
`@@ -45,15 +45,15 @@ func validateModelClaim(models []coreapi.OpenModel, playground inferenceapi.Pl`
`45`	`45`	`if playground.Spec.ModelClaim.ModelName != service.Spec.ModelClaims.Models[0].Name {`
`46`	`46`	`return fmt.Errorf("expected modelName %s, got %s", playground.Spec.ModelClaim.ModelName, service.Spec.ModelClaims.Models[0].Name)`
`47`	`47`	`}`
`48`		`- if diff := cmp.Diff(playground.Spec.ModelClaim.InferenceFlavorClaims, service.Spec.ModelClaims.InferenceFlavorClaims); diff != "" {`
`49`		`- return fmt.Errorf("unexpected flavors, want %v, got %v", playground.Spec.ModelClaim.InferenceFlavorClaims, service.Spec.ModelClaims.InferenceFlavorClaims)`
	`48`	`+ if diff := cmp.Diff(playground.Spec.ModelClaim.InferenceFlavors, service.Spec.ModelClaims.InferenceFlavors); diff != "" {`
	`49`	`+ return fmt.Errorf("unexpected flavors, want %v, got %v", playground.Spec.ModelClaim.InferenceFlavors, service.Spec.ModelClaims.InferenceFlavors)`
`50`	`50`	`}`
`51`	`51`	`} else if playground.Spec.ModelClaims != nil {`
`52`	`52`	`if diff := cmp.Diff(*playground.Spec.ModelClaims, service.Spec.ModelClaims); diff != "" {`
`53`	`53`	`return fmt.Errorf("expected modelClaims, want %v, got %v", *playground.Spec.ModelClaims, service.Spec.ModelClaims)`
`54`	`54`	`}`
`55`		`- if diff := cmp.Diff(playground.Spec.ModelClaims.InferenceFlavorClaims, service.Spec.ModelClaims.InferenceFlavorClaims); diff != "" {`
`56`		`- return fmt.Errorf("unexpected flavors, want %v, got %v", playground.Spec.ModelClaim.InferenceFlavorClaims, service.Spec.ModelClaims.InferenceFlavorClaims)`
	`55`	`+ if diff := cmp.Diff(playground.Spec.ModelClaims.InferenceFlavors, service.Spec.ModelClaims.InferenceFlavors); diff != "" {`
	`56`	`+ return fmt.Errorf("unexpected flavors, want %v, got %v", playground.Spec.ModelClaim.InferenceFlavors, service.Spec.ModelClaims.InferenceFlavors)`
`57`	`57`	`}`
`58`	`58`	`}`
`59`	`59`
Original file line number	Diff line number	Diff line change
`@@ -164,8 +164,8 @@ func ValidateModelLoader(model *coreapi.OpenModel, index int, template corev1.Po`
`164`	`164`
`165`	`165`	`func ValidateModelFlavor(service inferenceapi.Service, model coreapi.OpenModel, workload *lws.LeaderWorkerSet) error {`
`166`	`166`	`flavorName := model.Spec.InferenceConfig.Flavors[0].Name`
`167`		`- if len(service.Spec.ModelClaims.InferenceFlavorClaims) > 0 {`
`168`		`- flavorName = service.Spec.ModelClaims.InferenceFlavorClaims[0]`
	`167`	`+ if len(service.Spec.ModelClaims.InferenceFlavors) > 0 {`
	`168`	`+ flavorName = service.Spec.ModelClaims.InferenceFlavors[0]`
`169`	`169`	`}`
`170`	`170`
`171`	`171`	`for _, flavor := range model.Spec.InferenceConfig.Flavors {`
Original file line number	Diff line number	Diff line change
`@@ -67,7 +67,7 @@ func (w *PlaygroundWrapper) ModelClaim(modelName string, flavorNames ...string)`
`67`	`67`	`}`
`68`	`68`
`69`	`69`	`if len(names) > 0 {`
`70`		`- w.Spec.ModelClaim.InferenceFlavorClaims = names`
	`70`	`+ w.Spec.ModelClaim.InferenceFlavors = names`
`71`	`71`	`}`
`72`	`72`	`return w`
`73`	`73`	`}`
`@@ -87,7 +87,7 @@ func (w *PlaygroundWrapper) ModelClaims(modelNames []string, roles []string, fla`
`87`	`87`	`}`
`88`	`88`
`89`	`89`	`if len(fNames) > 0 {`
`90`		`- w.Spec.ModelClaims.InferenceFlavorClaims = fNames`
	`90`	`+ w.Spec.ModelClaims.InferenceFlavors = fNames`
`91`	`91`	`}`
`92`	`92`	`return w`
`93`	`93`	`}`
Original file line number	Diff line number	Diff line change
`@@ -59,7 +59,7 @@ func (w *ServiceWrapper) ModelClaims(modelNames []string, roles []string, flavor`
`59`	`59`	`}`
`60`	`60`
`61`	`61`	`if len(fNames) > 0 {`
`62`		`- w.Spec.ModelClaims.InferenceFlavorClaims = fNames`
	`62`	`+ w.Spec.ModelClaims.InferenceFlavors = fNames`
`63`	`63`	`}`
`64`	`64`	`return w`
`65`	`65`	`}`