init commit

clusters/k3s-dgx/kserve/gpu-serving-runtime.yaml

@@ -0,0 +1,42 @@
+apiVersion: serving.kserve.io/v1beta1
+kind: ServingRuntime
+metadata:
+  name: kserve-gpu-runtime
+  namespace: kserve
+spec:
+  supportedModelFormats:
+    - name: tensorflow
+      version: "2"
+      autoSelect: true
+    - name: pytorch
+      version: "2"
+      autoSelect: true
+    - name: sklearn
+      version: "1"
+      autoSelect: true
+    - name: xgboost
+      version: "1"
+      autoSelect: true
+  protocol: v1
+  protocolVersions:
+    - v1
+    - v2
+  containers:
+    - name: kserve-container
+      image: kserve/sklearnserver-gpu:latest
+      resources:
+        requests:
+          cpu: "1"
+          memory: "2Gi"
+          nvidia.com/gpu: "1"
+        limits:
+          cpu: "4"
+          memory: "8Gi"
+          nvidia.com/gpu: "1"
+      env:
+        - name: NVIDIA_VISIBLE_DEVICES
+          value: "all"
+        - name: NVIDIA_DRIVER_CAPABILITIES
+          value: "compute,utility"
+  multiModel: false
+  disabled: false

clusters/k3s-dgx/kserve/istio-gateway.yaml

@@ -0,0 +1,35 @@
+apiVersion: networking.istio.io/v1beta1
+kind: Gateway
+metadata:
+  name: kserve-gateway
+  namespace: kserve
+spec:
+  selector:
+    istio: ingressgateway
+  servers:
+    - port:
+        number: 80
+        name: http
+        protocol: HTTP
+      hosts:
+        - "*"
+---
+apiVersion: networking.istio.io/v1beta1
+kind: VirtualService
+metadata:
+  name: kserve-vs
+  namespace: kserve
+spec:
+  hosts:
+    - "*"
+  gateways:
+    - kserve-gateway
+  http:
+    - match:
+        - uri:
+            prefix: /v1/models/
+      route:
+        - destination:
+            host: kserve-default
+            port:
+              number: 80

clusters/k3s-dgx/kserve/kserve-controller.yaml

@@ -0,0 +1,43 @@
+apiVersion: source.toolkit.fluxcd.io/v1
+kind: HelmRepository
+metadata:
+  name: kserve
+  namespace: kserve
+spec:
+  interval: 10m
+  url: https://kserve.github.io/kserve
+---
+apiVersion: helm.toolkit.fluxcd.io/v2
+kind: HelmRelease
+metadata:
+  name: kserve
+  namespace: kserve
+spec:
+  interval: 10m
+  chart:
+    spec:
+      chart: kserve
+      version: "v0.12.0"
+      sourceRef:
+        kind: HelmRepository
+        name: kserve
+        namespace: kserve
+  values:
+    controller:
+      resources:
+        requests:
+          cpu: 500m
+          memory: 512Mi
+        limits:
+          cpu: 2
+          memory: 2Gi
+    config:
+      ingress:
+        className: istio
+      storage:
+        initialCapacity: 10Gi
+        storageClassName: local-path
+    knative:
+      enabled: true
+    istio:
+      enabled: true

clusters/k3s-dgx/kserve/kserve-namespace.yaml

@@ -0,0 +1,7 @@
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: kserve
+  labels:
+    istio-injection: enabled
+    serving.kserve.io/serving-runtime: "true"

clusters/k3s-dgx/kserve/kustomization.yaml

@@ -0,0 +1,10 @@
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+namespace: kserve
+resources:
+  - kserve-namespace.yaml
+  - kserve-controller.yaml
+  - istio-gateway.yaml
+  # - gpu-serving-runtime.yaml
+  # - model-storage-pvc.yaml
+  # - storage-config.yaml

clusters/k3s-dgx/kserve/model-storage-pvc.yaml

@@ -0,0 +1,12 @@
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: model-storage-pvc
+  namespace: kserve
+spec:
+  accessModes:
+    - ReadWriteOnce
+  resources:
+    requests:
+      storage: 50Gi
+  storageClassName: local-path

clusters/k3s-dgx/kserve/storage-config.yaml

@@ -0,0 +1,20 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: storage-config
+  namespace: kserve
+data:
+  storageConfig.yaml: |
+    defaultStorageUri: "pvc://model-storage"
+    storageSpec:
+      - name: model-storage
+        type: pvc
+        pvcSpec:
+          accessModes:
+            - ReadWriteOnce
+          resources:
+            requests:
+              storage: 50Gi
+          storageClassName: local-path
+    modelCacheSize: 10Gi
+    modelCacheMemory: 2Gi