apiVersion: apps/v1
kind: Deployment
metadata:
  name: a100-brain-vllm
  namespace: customer1
  labels:
    app: a100-brain
spec:
  replicas: 1
  selector:
    matchLabels:
      app: a100-brain 
  template:
    metadata:
      labels:
        app: a100-brain 
    spec:
      nodeSelector:
        cloud.google.com/gke-accelerator: "nvidia-a100-80gb"
      tolerations:
        - key: "nvidia.com/gpu-a100-80gb"
          operator: "Equal"
          value: "present"
          effect: "NoSchedule"
      containers:
        - name: vllm-brain
          image: vllm/vllm-openai:latest
          command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
          env:
            - name: HF_TOKEN
              value: ""
            - name: TOKENIZER_MODE
              value: "hf"
          args:
            - --model=sakamakismile/Huihui-Qwen3.6-27B-abliterated-NVFP4-MTP
            - --host=0.0.0.0
            - --port=8000
            - --tensor-parallel-size=1
            - --tokenizer-mode=hf 
            - --gpu-memory-utilization=0.95
            - --kv-cache-dtype=fp8  
            - --max-model-len=262144
            - --enable-auto-tool-choice 
            - --enable-chunked-prefill
            - --max-num-batched-tokens=16384 
            - --max-num-seqs=2
            - --trust-remote-code
            - --dtype=auto 
            - --enable-prefix-caching
            - --tool-call-parser=qwen3_xml
            - --reasoning-parser=qwen3
            - --disable-custom-all-reduce
            - --attention-backend=flash_attn
            - --quantization=modelopt 
          ports:
            - containerPort: 8000
          resources:
            limits:
              nvidia.com/gpu: 1
              
              
            requests:
              nvidia.com/gpu: 1
              memory: "100Gi" 
              cpu: "24" 
          
          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-4bit
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: vllm-model-qwen3.6-27b-uncensored-4bit
  namespace: customer1
spec:
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 100Gi
---
apiVersion: v1
kind: Service
metadata:
  name: a100-brain-service 
  namespace: customer1
spec:
  selector:
    app: a100-brain 
  ports:
    - protocol: TCP
      port: 8000
      targetPort: 8000
  type: ClusterIP