deploy.yaml

# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
  name: llama3-70b-disagg-sn
spec:
  backendFramework: vllm
  pvcs:
    - name: model-cache
      create: false
  services:
    Frontend:
      componentType: frontend
      dynamoNamespace: llama3-70b-disagg-sn
      volumeMounts:
        - name: model-cache
          mountPoint: /root/.cache/huggingface
      extraPodSpec:
        mainContainer:
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
          workingDir: /workspace/components/backends/vllm
      replicas: 1
    VllmPrefillWorker:
      componentType: worker
      dynamoNamespace: llama3-70b-disagg-sn
      envFromSecret: hf-token-secret
      volumeMounts:
        - name: model-cache
          mountPoint: /root/.cache/huggingface
      sharedMemory:
        size: 80Gi
      extraPodSpec:
        affinity:
          podAffinity:
            requiredDuringSchedulingIgnoredDuringExecution:
              - labelSelector:
                  matchExpressions:
                    - key: nvidia.com/dynamo-component-type
                      operator: In
                      values:
                        - worker
                topologyKey: kubernetes.io/hostname
        mainContainer:
          args:
          - "python3 -m dynamo.vllm --model RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic --tensor-parallel-size 2 --data-parallel-size 1 --disable-log-requests --is-prefill-worker --gpu-memory-utilization 0.95 --no-enable-prefix-caching --block-size 128"
          command:
          - /bin/sh
          - -c
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
          workingDir: /workspace/components/backends/vllm
      replicas: 2
      resources:
        limits:
          gpu: "2"
        requests:
          gpu: "2"
    VllmDecodeWorker:
      componentType: worker
      dynamoNamespace: llama3-70b-disagg-sn
      envFromSecret: hf-token-secret
      volumeMounts:
        - name: model-cache
          mountPoint: /root/.cache/huggingface
      sharedMemory:
        size: 80Gi
      extraPodSpec:
        affinity:
          podAffinity:
            requiredDuringSchedulingIgnoredDuringExecution:
              - labelSelector:
                  matchExpressions:
                    - key: nvidia.com/dynamo-component-type
                      operator: In
                      values:
                        - worker
                topologyKey: kubernetes.io/hostname
        mainContainer:
          args:
          - "python3 -m dynamo.vllm --model RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic --tensor-parallel-size 4 --data-parallel-size 1 --disable-log-requests --gpu-memory-utilization 0.90 --no-enable-prefix-caching --block-size 128"
          command:
          - /bin/sh
          - -c
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
          workingDir: /workspace/components/backends/vllm
      replicas: 1
      resources:
        limits:
          gpu: "4"
        requests:
          gpu: "4"