06-serving-deployment.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: qwen2.5-7b-k8s-chat
  namespace: llm-serving
spec:
  replicas: 2
  selector:
    matchLabels:
      app: qwen2.5-7b-k8s-chat
  template:
    metadata:
      labels:
        app: qwen2.5-7b-k8s-chat
    spec:
      nodeSelector:
        accelerator: nvidia-a100
      containers:
        - name: vllm-server
          image: registry.example.com/llm/vllm-openai:0.6.x-cu124
          args:
            - "--model"
            - "/models/runtime"
            - "--served-model-name"
            - "qwen2.5-7b-k8s-chat"
            - "--tensor-parallel-size"
            - "2"
            - "--max-model-len"
            - "8192"
            - "--max-num-seqs"
            - "32"
            - "--gpu-memory-utilization"
            - "0.92"
            - "--quantization"
            - "awq"
            - "--enable-prefix-caching"
            - "--port"
            - "8000"
          ports:
            - containerPort: 8000
              name: http
          resources:
            requests:
              cpu: "8"
              memory: "64Gi"
              nvidia.com/gpu: "2"
            limits:
              cpu: "8"
              memory: "64Gi"
              nvidia.com/gpu: "2"
          readinessProbe:
            httpGet:
              path: /health
              port: http
            initialDelaySeconds: 20
            periodSeconds: 10
          livenessProbe:
            httpGet:
              path: /health
              port: http
            initialDelaySeconds: 30
            periodSeconds: 15
          volumeMounts:
            - name: runtime-package
              mountPath: /models/runtime
      volumes:
        - name: runtime-package
          persistentVolumeClaim:
            claimName: llm-runtime-package-pvc
---
apiVersion: v1
kind: Service
metadata:
  name: qwen2.5-7b-k8s-chat
  namespace: llm-serving
spec:
  selector:
    app: qwen2.5-7b-k8s-chat
  ports:
    - name: http
      port: 80
      targetPort: 8000

# 这份清单想强调：
# - 线上部署对象关注副本、GPU、探针和挂载。
# - 它不是模型训练产物，也不是 model registry 条目。