kserve/config/runtimes/kserve-tensorflow-serving.yaml

apiVersion: serving.kserve.io/v1alpha1
kind: ClusterServingRuntime
metadata:
  name: kserve-tensorflow-serving
  annotations:
    serving.kserve.io/server-type: tensorflow-serving
spec:
  annotations:
    prometheus.kserve.io/port: '8080'
    prometheus.kserve.io/path: "/metrics"
  supportedModelFormats:
    - name: tensorflow
      version: "1"
      autoSelect: true
      priority: 2
    - name: tensorflow
      version: "2"
      autoSelect: true
      priority: 2
  protocolVersions:
    - v1
    - grpc-v1
  containers:
    - name: kserve-container
      image: tensorflow-serving:replace
      command: [/usr/bin/tensorflow_model_server]
      args:
        - --model_name={{.Name}}
        - --port=9000
        - --rest_api_port=8080
        - --model_base_path=/mnt/models
        - --rest_api_timeout_in_ms=60000
      securityContext:
        runAsUser: 1000    # User is not defined in the Dockerfile, so we need to set it here to run as non-root
        allowPrivilegeEscalation: false
        privileged: false
        runAsNonRoot: true
        capabilities:
          drop:
            - ALL
      resources:
        requests:
          cpu: "1"
          memory: 2Gi
        limits:
          cpu: "1"
          memory: 2Gi