apiVersion: serving.kserve.io/v1alpha1
kind: LLMInferenceService
metadata:
  name: huihui-granite
spec:
  model:
    modelFormat:
      name: huggingface
    storageUri: "hf://huihui-ai/Huihui-granite-4.1-30b-abliterated"
  replicas: 1
  template:
    containers:
      - name: main
        image: quay.io/pierdipi/vllm-cpu:latest
        args:
          - --model_name=huihui-granite
        securityContext:
          runAsNonRoot: false
        env:
          - name: VLLM_LOGGING_LEVEL
            value: DEBUG
        resources:
          limits:
            cpu: "4"
            memory: 16Gi
            nvidia.com/gpu: "1"
          requests:
            cpu: "2"
            memory: 8Gi
            nvidia.com/gpu: "1"
        livenessProbe:
          initialDelaySeconds: 30
          periodSeconds: 30
          timeoutSeconds: 30
          failureThreshold: 5
  router:
    gateway: {}
    route: {}
    scheduler: {}