fix(modeldeployment): use v1beta1 InferenceSet and allow empty instanceType

charts/modeldeployment/README.md

@@ -3,7 +3,7 @@
 Helm chart that deploys a complete set of GAIE (Gateway API Inference
 Extension) artifacts for a single model:
 
-- `kaito.sh/v1alpha1` `InferenceSet`
+- `kaito.sh/v1beta1` `InferenceSet`
 - `inference.networking.k8s.io/v1` `InferencePool` (normally provisioned by KAITO via Flux; rendered inline here)
 - Endpoint Picker (EPP) `Deployment` + `Service` + `ServiceAccount` + `Role` + `RoleBinding` + `ConfigMap`
 - `gateway.networking.k8s.io/v1` `HTTPRoute` matching `X-Gateway-Model-Name: <name>`
@@ -41,7 +41,7 @@ over plaintext gRPC and **no `DestinationRule` is required**.
 | `model`                   | required | `""`                                                                 | Preset model name. Used only as `spec.template.inference.preset.name`.                     |
 | `config`                  | optional | `""`                                                                 | Name of a ConfigMap in the deployment namespace containing custom inference arguments. Rendered as `spec.template.inference.config`. |
 | `modelAccessSecret`       | optional | `""`                                                                 | Name of a Secret containing Hugging Face model access credentials. Rendered as `spec.template.inference.preset.presetOptions.modelAccessSecret`. |
-| `instanceType`            | required | `Standard_NV36ads_A10_v5`                                            | VM instance type for the underlying nodes.                                                 |
+| `instanceType`            | optional | `Standard_NV36ads_A10_v5`                                            | VM instance type for the underlying nodes. Set `""` to omit `spec.template.resource.instanceType`, as required on BYO-node clusters (KAITO `nodeProvisioner=byo`). |
 | `replicas`                | optional | _empty_                                                              | InferenceSet replicas (`spec.replicas`). Empty leaves `spec.replicas` unset (CRD default `1`; with `enableScaling=true`, KEDA moves it into `[minReplicas, maxReplicas]`). When set with `enableScaling=true` it must lie within `[minReplicas, maxReplicas]` and is re-applied only when its value changes (see [Replica ownership](#replica-ownership)). |
 | `minReplicas`             | required when `enableScaling=true` | _empty_                                    | Autoscaling lower bound, wired to `scaledobject.kaito.sh/min-replicas`. Ignored when `enableScaling=false`. `0` enables [scale to zero](#scale-to-zero). |
 | `enableScaling`           | optional | `false`                                                              | Wired to `scaledobject.kaito.sh/auto-provision`. Gates `minReplicas`, `maxReplicas` and the entire `scaling` block. |

charts/modeldeployment/templates/inferenceset.yaml

@@ -130,7 +130,7 @@ install time instead of surfacing as a skipped ScaledObject.
 {{- end }}
 {{- end }}
 {{- end }}
-apiVersion: kaito.sh/v1alpha1
+apiVersion: kaito.sh/v1beta1
 kind: InferenceSet
 metadata:
   name: {{ $name }}
@@ -232,8 +232,15 @@ spec:
         inference.kaito.sh/stream-source-type: {{ .Values.streaming.sourceType | quote }}
         {{- end }}
       {{- end }}
+    {{- /*
+    Empty instanceType omits the field, which KAITO requires on BYO
+    (nodeProvisioner=byo) clusters where nodes are not auto-provisioned.
+    */}}
     resource:
-      instanceType: {{ .Values.instanceType | quote }}
+      {{- with .Values.instanceType }}
+      instanceType: {{ . | quote }}
+      {{- else }} {}
+      {{- end }}
     inference:
       preset:
         name: {{ .Values.model | quote }}

charts/modeldeployment/values.schema.json

@@ -85,8 +85,7 @@
     },
     "instanceType": {
       "type": "string",
-      "minLength": 1,
-      "description": "VM instance type used for the underlying GPU nodes."
+      "description": "VM instance type used for the underlying GPU nodes. Empty omits spec.template.resource.instanceType, as required on BYO-node clusters (KAITO nodeProvisioner=byo)."
     },
     "replicas": {
       "anyOf": [

charts/modeldeployment/values.yaml

@@ -16,7 +16,7 @@ config: ""
 # modelAccessSecret names the Secret holding the Hugging Face access credentials.
 modelAccessSecret: ""
 
-# instanceType is the VM instance type of the GPU nodes.
+# instanceType is the VM instance type of the GPU nodes; set "" on BYO-node clusters to omit it.
 instanceType: "Standard_NV36ads_A10_v5"
 
 # replicas sets spec.replicas when not empty (otherwise the CRD default of 1); with enableScaling it must lie

test/e2e/modeldeployment_chart_spec.go

@@ -102,7 +102,7 @@ var _ = Describe("ModelDeployment Chart", utils.GinkgoLabelInferenceSet, utils.G
 			is := &unstructured.Unstructured{}
 			is.SetGroupVersionKind(schema.GroupVersionKind{
 				Group:   "kaito.sh",
-				Version: "v1alpha1",
+				Version: "v1beta1",
 				Kind:    "InferenceSet",
 			})
 			err := cl.Get(ctx, types.NamespacedName{Name: deploymentName, Namespace: namespace}, is)
@@ -187,7 +187,7 @@ var _ = Describe("ModelDeployment Chart", utils.GinkgoLabelInferenceSet, utils.G
 				is := &unstructured.Unstructured{}
 				is.SetGroupVersionKind(schema.GroupVersionKind{
 					Group:   "kaito.sh",
-					Version: "v1alpha1",
+					Version: "v1beta1",
 					Kind:    "InferenceSet",
 				})
 				err := cl.Get(ctx, types.NamespacedName{Name: deploymentName, Namespace: namespace}, is)
@@ -246,7 +246,7 @@ var _ = Describe("ModelDeployment Chart", utils.GinkgoLabelInferenceSet, utils.G
 			is := &unstructured.Unstructured{}
 			is.SetGroupVersionKind(schema.GroupVersionKind{
 				Group:   "kaito.sh",
-				Version: "v1alpha1",
+				Version: "v1beta1",
 				Kind:    "InferenceSet",
 			})
 			err := cl.Get(ctx, types.NamespacedName{Name: deploymentName, Namespace: namespace}, is)

test/e2e/utils/dynamic.go

@@ -25,7 +25,7 @@ var (
 	// InferenceSetGVR is the GroupVersionResource for KAITO InferenceSet objects.
 	InferenceSetGVR = schema.GroupVersionResource{
 		Group:    "kaito.sh",
-		Version:  "v1alpha1",
+		Version:  "v1beta1",
 		Resource: "inferencesets",
 	}