disagg.yaml 1.42 KB
Newer Older
1
2
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
Alec's avatar
Alec committed
3

4
5
6
apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
7
  name: vllm-disagg
8
9
10
spec:
  services:
    Frontend:
11
      dynamoNamespace: vllm-disagg
12
      componentType: frontend
13
14
15
      replicas: 1
      extraPodSpec:
        mainContainer:
16
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
17
    VllmDecodeWorker:
18
      dynamoNamespace: vllm-disagg
19
      envFromSecret: hf-token-secret
20
      componentType: worker
21
22
23
      replicas: 1
      resources:
        limits:
24
          gpu: "1"
25
26
      extraPodSpec:
        mainContainer:
27
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
Alec's avatar
Alec committed
28
          workingDir: /workspace/components/backends/vllm
29
30
31
          command:
            - /bin/sh
            - -c
32
          args:
33
            - "python3 -m dynamo.vllm --model Qwen/Qwen3-0.6B"
34
    VllmPrefillWorker:
35
      dynamoNamespace: vllm-disagg
36
      envFromSecret: hf-token-secret
37
      componentType: worker
38
39
40
      replicas: 1
      resources:
        limits:
41
          gpu: "1"
42
43
      extraPodSpec:
        mainContainer:
44
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
Alec's avatar
Alec committed
45
          workingDir: /workspace/components/backends/vllm
46
47
48
          command:
            - /bin/sh
            - -c
49
          args:
50
            - "python3 -m dynamo.vllm --model Qwen/Qwen3-0.6B --is-prefill-worker"