---
apiVersion: serving.kserve.io/v1alpha2
kind: LLMInferenceServiceConfig
metadata:
  annotations:
    internal.config.kubernetes.io/previousKinds: LLMInferenceServiceConfig
    internal.config.kubernetes.io/previousNames: kserve-config-llm-router-route
    internal.config.kubernetes.io/previousNamespaces: opendatahub
    serving.kserve.io/well-known-config: "true"
  creationTimestamp: "2026-11-03T18:46:30Z"
  finalizers:
  - serving.kserve.io/llmisvcconfig-finalizer
  generation: 2
  labels:
    platform.opendatahub.io/part-of: kserve
  managedFields:
  - apiVersion: serving.kserve.io/v1alpha2
    fieldsType: FieldsV1
    fieldsV1:
      f:metadata:
        f:annotations:
          f:internal.config.kubernetes.io/previousKinds: {}
          f:internal.config.kubernetes.io/previousNames: {}
          f:internal.config.kubernetes.io/previousNamespaces: {}
          f:serving.kserve.io/well-known-config: {}
        f:labels:
          f:platform.opendatahub.io/part-of: {}
      f:spec:
        f:router:
          f:route:
            f:http:
              f:spec:
                f:parentRefs: {}
                f:rules: {}
    manager: kserve-module-controller
    operation: Apply
    time: "2026-11-03T18:46:30Z"
  - apiVersion: serving.kserve.io/v1alpha2
    fieldsType: FieldsV1
    fieldsV1:
      f:metadata:
        f:finalizers:
          .: {}
          v:"serving.kserve.io/llmisvcconfig-finalizer": {}
      f:spec:
        f:model:
          .: {}
          f:uri: {}
    manager: manager
    operation: Update
    time: "2026-11-03T18:46:31Z"
  - apiVersion: serving.kserve.io/v1alpha2
    fieldsType: FieldsV1
    fieldsV1:
      f:status:
        .: {}
        f:conditions: {}
        f:observedGeneration: {}
        f:referencedBy: {}
    manager: manager
    operation: Update
    subresource: status
    time: "2026-11-03T18:48:03Z"
  name: v3-6-0-ea-2-kserve-config-llm-router-route
  namespace: opendatahub
  resourceVersion: "19975"
  uid: efb26e7f-3e7e-4624-a8e2-bc1405d02720
spec:
  model:
    uri: ""
  router:
    route:
      http:
        spec:
          parentRefs:
          - group: gateway.networking.k8s.io
            kind: Gateway
            name: '{{ .GlobalConfig.IngressGatewayName }}'
            namespace: '{{ .GlobalConfig.IngressGatewayNamespace }}'
          rules:
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /v1/completions
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /{{ .ObjectMeta.Namespace }}/{{ .ObjectMeta.Name }}/v1/completions
            name: v1-completions-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /v1/chat/completions
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /{{ .ObjectMeta.Namespace }}/{{ .ObjectMeta.Name }}/v1/chat/completions
            name: v1-chat-completions-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /v1/responses
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /{{ .ObjectMeta.Namespace }}/{{ .ObjectMeta.Name }}/v1/responses
            name: v1-responses-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /v1/messages
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /{{ .ObjectMeta.Namespace }}/{{ .ObjectMeta.Name }}/v1/messages
            name: v1-messages-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            matches:
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: Exact
                value: /v1/completions
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: Exact
                value: /v1/completions/
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: Exact
                value: /v1/chat/completions
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: Exact
                value: /v1/chat/completions/
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: Exact
                value: /v1/responses
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: Exact
                value: /v1/responses/
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: Exact
                value: /v1/messages
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: Exact
                value: /v1/messages/
            name: v1-model-routing
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /v1/completions
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}/v1/completions
            name: v1-completions-publisher-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /v1/chat/completions
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}/v1/chat/completions
            name: v1-chat-completions-publisher-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /v1/responses
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}/v1/responses
            name: v1-responses-publisher-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: inference.networking.k8s.io
              kind: InferencePool
              name: '{{ ChildName .ObjectMeta.Name `-inference-pool` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /v1/messages
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}/v1/messages
            name: v1-messages-publisher-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: ""
              kind: Service
              name: '{{ ChildName .ObjectMeta.Name `-kserve-workload-svc` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
            name: v1-publisher-path-catch-all
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: ""
              kind: Service
              name: '{{ ChildName .ObjectMeta.Name `-kserve-workload-svc` }}'
              port: 8000
              weight: 1
            filters:
            - type: URLRewrite
              urlRewrite:
                path:
                  replacePrefixMatch: /
                  type: ReplacePrefixMatch
            matches:
            - path:
                type: PathPrefix
                value: /{{ .ObjectMeta.Namespace }}/{{ .ObjectMeta.Name }}
            name: v1-catch-all-path
            timeouts:
              backendRequest: 0s
              request: 0s
          - backendRefs:
            - group: ""
              kind: Service
              name: '{{ ChildName .ObjectMeta.Name `-kserve-workload-svc` }}'
              port: 8000
              weight: 1
            matches:
            - headers:
              - name: '{{ .GlobalConfig.ModelBasedRoutingHeaderName }}'
                type: Exact
                value: publishers/{{ .ObjectMeta.Namespace }}/models/{{ .Spec.Model.Name
                  }}
              path:
                type: PathPrefix
                value: /
            name: v1-catch-all-model-routing
            timeouts:
              backendRequest: 0s
              request: 0s
status:
  conditions:
  - lastTransitionTime: "2026-11-03T18:48:03Z"
    message: referenced by 8 LLMInferenceService(s)
    reason: InUse
    severity: Info
    status: "True"
    type: ConfigInUse
  - lastTransitionTime: "2026-11-03T18:46:31Z"
    status: "True"
    type: Ready
  observedGeneration: 2
  referencedBy:
  - name: e2e-distinct-2-simulated
    namespace: llm
  - name: e2e-distinct-simulated
    namespace: llm
  - name: e2e-embedding-simulated
    namespace: llm
  - name: e2e-trlp-test-simulated
    namespace: llm
  - name: e2e-unconfigured-facebook-opt-125m-simulated
    namespace: llm
  - name: echo-simulator
    namespace: llm
  - name: facebook-opt-125m-simulated
    namespace: llm
  - name: premium-simulated-simulated-premium
    namespace: llm