disagg_router.yaml 1.54 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
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/nvidian/nim-llm-dev/vllm-runtime:dep-233.17
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/nvidian/nim-llm-dev/vllm-runtime:dep-233.17
Alec's avatar
Alec committed
31
          workingDir: /workspace/components/backends/vllm
32
33
34
          command:
            - /bin/sh
            - -c
35
          args:
36
            - python3 -m dynamo.vllm --model Qwen/Qwen3-0.6B
37
38
39
40
41
    VllmPrefillWorker:
      dynamoNamespace: vllm-v1-disagg-router
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 1
42
43
44
      resources:
        limits:
          gpu: "1"
45
46
      extraPodSpec:
        mainContainer:
47
          image: nvcr.io/nvidian/nim-llm-dev/vllm-runtime:dep-233.17
Alec's avatar
Alec committed
48
          workingDir: /workspace/components/backends/vllm
49
50
51
          command:
            - /bin/sh
            - -c
52
          args:
53
            - python3 -m dynamo.vllm --model Qwen/Qwen3-0.6B --is-prefill-worker