tutorials-and-examples/nvidia-bionemo/fine-tuning/inference/deployment.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  namespace: bionemo-training
  name: esm2-inference
  labels:
    app: esm2-inference
spec:
  replicas: 1
  selector:
    matchLabels:
      app: esm2-inference
  template:
    metadata:
      labels:
        app: esm2-inference
    spec:
      serviceAccountName: esm2-inference-sa
      containers:
      - name: inference
        image: esm2-inference-image
        ports:
        - containerPort: 8000
          name: http
        env:
        - name: MODEL_PATH
          value: "/mnt/data/model"
        resources:
          limits:
            nvidia.com/gpu: 1
            memory: "4Gi"
            cpu: "2"
          requests:
            nvidia.com/gpu: 1
            memory: "2Gi"
            cpu: "500m"
        volumeMounts:
        - name: model-storage
          mountPath: /mnt/data
          readOnly: true
        livenessProbe:
          httpGet:
            path: /health
            port: 8000
          initialDelaySeconds: 30
          periodSeconds: 10
        readinessProbe:
          httpGet:
            path: /health
            port: 8000
          initialDelaySeconds: 15
          periodSeconds: 5
        securityContext:
          runAsNonRoot: true
          runAsUser: 1000
          allowPrivilegeEscalation: false
      volumes:
      - name: model-storage
        persistentVolumeClaim:
          claimName: bionemo-filestore
          readOnly: true
      nodeSelector:
        cloud.google.com/gke-gpu: "true"
---
apiVersion: v1
kind: Service
metadata:
  name: esm2-inference
spec:
  selector:
    app: esm2-inference
  ports:
  - port: 80
    targetPort: 8000
    protocol: TCP
    name: http
  type: ClusterIP