apiVersion: apps/v1
kind: Deployment
metadata:
  name: server
spec:
  replicas: 1
  strategy:
    type: Recreate
  selector:
    matchLabels:
      app.kubernetes.io/component: server
  template:
    metadata:
      labels:
        app.kubernetes.io/component: server
    spec:
      affinity:
        nodeAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
            nodeSelectorTerms:
              - matchExpressions:
                  # Image only supports amd64
                  - key: kubernetes.io/arch
                    operator: In
                    values:
                      - amd64
      priorityClassName: high-priority
      runtimeClassName: nvidia
      containers:
        - name: server
          image: ollama/ollama
          # image: badjware/ollama-tweak
          env:
            - name: OLLAMA_KEEP_ALIVE
              value: 36h
            # - name: HSA_OVERRIDE_GFX_VERSION
            #   value: 10.1.0
            # - name: HSA_ENABLE_SDMA
            #   value: "0"
          resources:
            requests:
              memory: 4Gi
              cpu: 2000m
            limits:
              memory: 4Gi
              cpu: 3000m
              nvidia.com/gpu: 1
          ports:
            - containerPort: 11434
              name: http
          volumeMounts:
            - name: server-data
              mountPath: /root/.ollama
      volumes:
        - name: server-data
          hostPath:
            path: /var/lib/ollama
            type: DirectoryOrCreate
---
apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: high-priority
value: 1000000
globalDefault: false
---
apiVersion: v1
kind: Service
metadata:
  name: server
  labels:
    app.kubernetes.io/component: server
spec:
  selector:
    app.kubernetes.io/component: server
  ports:
    - name: http
      port: 11434
      targetPort: http