sd21-512-serve-deploy.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: sd21-inf2-serve
  name: sd21-inf2-serve
spec:
  selector:
    matchLabels:
      app: sd21-inf2-serve
  template:
    metadata:
      labels:
        app: sd21-inf2-serve
    spec:
      nodeSelector:
        karpenter.sh/nodepool: inf2
      serviceAccountName: sd21-sa
      schedulerName: my-scheduler
      volumes:
      - name: dshm
        emptyDir:
          medium: Memory
      initContainers:
      - name: pull-model-tar
        image: ${AWS_ACCOUNT_ID}.dkr.ecr.${AWS_REGION}.amazonaws.com/stablediffusion:amd64-neuron-assets
        imagePullPolicy: Always
        volumeMounts:
        - name: workdir
          mountPath: /model
        command:
        - /bin/bash
        - "-exc"
        - |
          set -x
          aws s3 cp s3://sdinfer/stable-diffusion-2-1-base.tar.gz /model/model.tar.gz
          cd /model
          tar -xzf /model/model.tar.gz
          ls /model
      containers:
      - name: app
        image: ${AWS_ACCOUNT_ID}.dkr.ecr.${AWS_REGION}.amazonaws.com/stablediffusion:neuron2.19
        imagePullPolicy: Always
        volumeMounts:
          - mountPath: /dev/shm
            name: dshm
          - mountPath: /app
            name: workdir
        command: ["/start.sh"]
        args: ["run"]
        resources:
          limits:
            aws.amazon.com/neuron: 2
        env:
        - name: DEVICE
          value: "xla"
        - name: BUCKET
          value: "sdinfer"
        - name: MODEL_ID
          value: "stabilityai/stable-diffusion-2-1-base"
        - name: COMPILER_WORKDIR_ROOT
          value: "/app/sd2_compile_dir_512"
        - name: MODEL_FILE
          value: "stable-diffusion-2-1-base"
        - name: POD_NAME
          valueFrom:
            fieldRef:
              fieldPath: metadata.name
        ports:
          - containerPort: 8000
            protocol: TCP
        readinessProbe:
          httpGet:
            path: /readiness
            port: 8000
          initialDelaySeconds: 60
          periodSeconds: 10
      volumes:
      - name: workdir
        emptyDir: {}
      - name: dshm
        emptyDir:
          medium: Memory