apiVersion: apps/v1
kind: Deployment
metadata:
  name: rtx6000-brain-vllm
  namespace: customer1
  labels:
    app: rtx6000-brain
spec:
  replicas: 1
  selector:
    matchLabels:
      app: rtx6000-brain
  template:
    metadata:
      labels:
        app: rtx6000-brain
    spec:
      nodeSelector:
        cloud.google.com/gke-accelerator: "nvidia-rtx-pro-6000"
      tolerations:
        - key: "nvidia.com/gpu-nvidia-rtx-pro-6000"
          operator: "Equal"
          value: "present"
          effect: "NoSchedule"
      containers:
        - name: vllm-brain
          image: vllm/vllm-openai:latest-cu129-ubuntu2404 
          command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
          env:
            - name: HF_TOKEN
              value: ""
            - name: TOKENIZER_MODE
              value: "hf"
          args:
            - --model=edp1096/Huihui-Qwen3.6-27B-abliterated-FP8
            - --host=0.0.0.0
            - --port=8000
            - --tensor-parallel-size=1
            - --tokenizer-mode=hf 
            - --gpu-memory-utilization=0.95
            - --max-model-len=98304
            - --enable-auto-tool-choice 
            - --kv-cache-dtype=fp8 
            - --max-num-batched-tokens=65536 
            - --max-num-seqs=128
            - --enable-chunked-prefill
            - --trust-remote-code
            - --dtype=auto
            - --enable-prefix-caching
            - --tool-call-parser=qwen3_xml
            - --reasoning-parser=qwen3
            
          ports:
            - containerPort: 8000
          resources:
            limits:
              nvidia.com/gpu: 1
              
              
            requests:
              nvidia.com/gpu: 1
              memory: "150Gi"
              cpu: "32"
          
          startupProbe:
            httpGet:
              path: /health
              port: 8000
            initialDelaySeconds: 30
            periodSeconds: 10
            timeoutSeconds: 10
            failureThreshold: 60
            successThreshold: 1 
          
          livenessProbe:
            httpGet:
              path: /health
              port: 8000 
            initialDelaySeconds: 180
            periodSeconds: 15
            failureThreshold: 5
            successThreshold: 1 
            timeoutSeconds: 10 
            
          
          readinessProbe:
            httpGet:
              path: /v1/models 
              port: 8000
            initialDelaySeconds: 120
            periodSeconds: 10 
            timeoutSeconds: 15 
            failureThreshold: 8
            successThreshold: 1 
          
          volumeMounts:
            - name: model-cache
              mountPath: /root/.cache/huggingface
      volumes:
        - name: model-cache
          persistentVolumeClaim:
            claimName: vllm-model-qwen3.6-27b-uncensored
---
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: hyperdisk-balanced
provisioner: pd.csi.storage.gke.io
volumeBindingMode: WaitForFirstConsumer
allowVolumeExpansion: true
parameters:
  type: hyperdisk-balanced
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: vllm-model-qwen3.6-27b-uncensored
  namespace: customer1
spec:
  accessModes:
    - ReadWriteOnce
  storageClassName: hyperdisk-balanced   
  resources:
    requests:
      storage: 100Gi
  
---
apiVersion: v1
kind: Service
metadata:
  name: rtx6000-brain-service
  namespace: customer1
  annotations:
    tailscale.com/proxy-group: rtx6000-brain
spec:
  selector:
    app: rtx6000-brain
  ports:
    - protocol: TCP
      port: 8000
      targetPort: 8000
  type: ClusterIP