use-cases/model-fine-tuning-pipeline/model-eval/manifests/deployment-l4.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: vllm-openai-l4
spec:
  replicas: 1
  selector:
    matchLabels:
      app: vllm-openai-l4
  template:
    metadata:
      labels:
        app: vllm-openai-l4
        ml-platform: vllm-openai
      annotations:
        gke-gcsfuse/volumes: "true"
    spec:
      containers:
      - name: inference-server
        args:
        - --model=$(MODEL)
        - --tensor-parallel-size=2
        env:
        - name: MODEL
          value: V_MODEL_PATH
        - name: VLLM_ATTENTION_BACKEND
          value: FLASHINFER
        image: V_IMAGE_URL
        readinessProbe:
          failureThreshold: 3
          httpGet:
            path: /health
            port: 8000
            scheme: HTTP
          initialDelaySeconds: 240
          periodSeconds: 5
          successThreshold: 1
          timeoutSeconds: 1
        resources:
          requests:
            cpu: "2"
            memory: "25Gi"
            ephemeral-storage: "25Gi"
            nvidia.com/gpu: "2"
          limits:
            cpu: "2"
            memory: "25Gi"
            ephemeral-storage: "25Gi"
            nvidia.com/gpu: "2"
        volumeMounts:
        - mountPath: /dev/shm
          name: dshm
        - name: gcs-fuse-csi-ephemeral
          mountPath: /model-data
          readOnly: true
      nodeSelector:
        cloud.google.com/gke-accelerator: nvidia-l4
      serviceAccountName: V_KSA
      tolerations:
      - key: "nvidia.com/gpu"
        operator: "Exists"
        effect: "NoSchedule"
      - key: "on-demand"
        value: "true"
        operator: "Equal"
        effect: "NoSchedule" 
      volumes:
      - name: dshm
        emptyDir:
            medium: Memory
      - name: gcs-fuse-csi-ephemeral
        csi:
          driver: gcsfuse.csi.storage.gke.io
          volumeAttributes:
            bucketName: V_BUCKET
            mountOptions: "implicit-dirs,file-cache:enable-parallel-downloads:true,file-cache:max-parallel-downloads:-1" 
            fileCacheCapacity: "20Gi"
            fileCacheForRangeRead: "true"
            metadataStatCacheCapacity: "-1"
            metadataTypeCacheCapacity: "-1"
            metadataCacheTTLSeconds: "-1"
---
apiVersion: v1
kind: Service
metadata:
  name: vllm-openai-l4
spec:
  selector:
    app: vllm-openai-l4
  type: ClusterIP
  ports:
    - protocol: TCP
      port: 8000
      targetPort: 8000