add class back

clusters/k3s-dgx/apps/huihui-granite-inference.yaml

@@ -1,18 +1,22 @@
-apiVersion: "serving.kserve.io/v1beta1"
+apiVersion: serving.kserve.io/v1beta1
-kind: "InferenceService"
+kind: InferenceService
 metadata:
-  name: "flower-sample"
+  name: huihui-granite
   namespace: kserve
 spec:
   predictor:
     model:
       modelFormat:
-        name: tensorflow
+        name: huggingface
-      storageUri: "gs://kfserving-examples/models/tensorflow/flowers"
+      args:
+        - --model_name=huihui-granite
+      storageUri: "hf://huihui-ai/Huihui-granite-4.1-30b-abliterated"
       resources:
-        requests:
-          cpu: "100m"
-          memory: "256Mi"
         limits:
-          cpu: "1"
+          cpu: "4"
-          memory: "1Gi"
+          memory: 16Gi
+          nvidia.com/gpu: "1"
+        requests:
+          cpu: "2"
+          memory: 8Gi
+          nvidia.com/gpu: "1"

clusters/k3s-dgx/infrastructure/kustomization.yaml

@@ -3,4 +3,4 @@ kind: Kustomization
 resources:
   - cert-manager.yaml
   - envoy-gateway-system.yaml
-  # - envoy-gateway-class.yaml
+  - envoy-gateway-class.yaml

clusters/k3s-dgx/kustomization.yaml

@@ -4,5 +4,5 @@ resources:
   - flux-system
   - gpu-support
   - infrastructure
-  # - kserve
+  - kserve
   # - apps