disagg_router.yaml 1.58 KB
Newer Older
1
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2
# SPDX-License-Identifier: Apache-2.0
Alec's avatar
Alec committed
3

4
5
6
7
8
9
10
11
apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
  name: vllm-v1-disagg-router
spec:
  services:
    Frontend:
      dynamoNamespace: vllm-v1-disagg-router
12
      componentType: frontend
13
14
15
      replicas: 1
      extraPodSpec:
        mainContainer:
16
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
17
18
19
      envs:
        - name: DYN_ROUTER_MODE
          value: kv
20
21
22
23
24
    VllmDecodeWorker:
      dynamoNamespace: vllm-v1-disagg-router
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 2
25
26
27
      resources:
        limits:
          gpu: "1"
28
29
      extraPodSpec:
        mainContainer:
30
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
31
          workingDir: /workspace/examples/backends/vllm
32
          command:
33
34
35
          - python3
          - -m
          - dynamo.vllm
36
          args:
37
38
            - --model
            - Qwen/Qwen3-0.6B
39
            - --is-decode-worker
40
41
42
43
    VllmPrefillWorker:
      dynamoNamespace: vllm-v1-disagg-router
      envFromSecret: hf-token-secret
      componentType: worker
44
      replicas: 2
45
46
47
      resources:
        limits:
          gpu: "1"
48
49
      extraPodSpec:
        mainContainer:
50
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
51
          workingDir: /workspace/examples/backends/vllm
52
          command:
53
54
55
          - python3
          - -m
          - dynamo.vllm
56
          args:
57
58
59
            - --model
            - Qwen/Qwen3-0.6B
            - --is-prefill-worker