"tests/planner/scaling/disagg_planner_load.yaml" did not exist on "28546bad2602e67eb9436de6198776855ba08040"
disagg.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
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:my-tag
17
    VllmDecodeWorker:
18
      dynamoNamespace: vllm-disagg
19
      envFromSecret: hf-token-secret
20
      componentType: worker
21
      subComponentType: decode
22
23
24
      replicas: 1
      resources:
        limits:
25
          gpu: "1"
26
27
      extraPodSpec:
        mainContainer:
28
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
29
          workingDir: /workspace/examples/backends/vllm
30
          command:
31
32
33
          - python3
          - -m
          - dynamo.vllm
34
          args:
35
36
            - --model
            - Qwen/Qwen3-0.6B
37
            - --is-decode-worker
38
    VllmPrefillWorker:
39
      dynamoNamespace: vllm-disagg
40
      envFromSecret: hf-token-secret
41
      componentType: worker
42
      subComponentType: prefill
43
44
45
      replicas: 1
      resources:
        limits:
46
          gpu: "1"
47
48
      extraPodSpec:
        mainContainer:
49
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
50
          workingDir: /workspace/examples/backends/vllm
51
          command:
52
53
54
          - python3
          - -m
          - dynamo.vllm
55
          args:
56
57
58
            - --model
            - Qwen/Qwen3-0.6B
            - --is-prefill-worker