# NOTE: this template does not really do any fancy node mapping or affinity declarations
# so the inference and safety models may land on the same GPU node
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: vllm-models
spec:
  accessModes:
    - ReadWriteOnce
  volumeMode: Filesystem
  storageClassName: gp2
  resources:
    requests:
      storage: 50Gi
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: vllm-server
spec:
  replicas: 1
  selector:
    matchLabels:
      app.kubernetes.io/name: vllm
  template:
    metadata:
      labels:
        app.kubernetes.io/name: vllm
    spec:
      containers:
      - name: vllm
        image: vllm/vllm-openai:latest
        command: ["/bin/sh", "-c"]
        args:
        - "vllm serve ${INFERENCE_MODEL} --dtype float16 --enforce-eager --max-model-len 4096 --gpu-memory-utilization 0.5"
        env:
        - name: HUGGING_FACE_HUB_TOKEN
          valueFrom:
            secretKeyRef:
              name: hf-token-secret
              key: token
        ports:
          - containerPort: 8000
        volumeMounts:
          - name: llama-storage
            mountPath: /root/.cache/huggingface
      volumes:
      - name: llama-storage
        persistentVolumeClaim:
          claimName: vllm-models