disagg-multinode.yaml 2.03 KB
Newer Older
1
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2
3
4
5
6
7
8
9
10
11
12
13
14
15
# SPDX-License-Identifier: Apache-2.0

apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
  name: vllm-disagg
spec:
  services:
    Frontend:
      componentType: frontend
      replicas: 1
      extraPodSpec:
        mainContainer:
          image: my-registry/vllm-runtime:my-tag
16
          workingDir: /workspace/examples/backends/vllm
17
          command:
18
19
20
          - python3
          - -m
          - dynamo.frontend
21
          args:
22
23
            - --http-port
            - "8000"
24
25
26
27
28
29
30
31
32
33
34
35
    decode:
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 1
      multinode:
        nodeCount: 2
      resources:
        limits:
          gpu: "1"
      extraPodSpec:
        mainContainer:
          image: my-registry/vllm-runtime:my-tag
36
          workingDir: /workspace/examples/backends/vllm
37
          command:
38
39
40
          - python3
          - -m
          - dynamo.vllm
41
          args:
42
43
44
45
46
47
48
49
          - --model
          - Qwen/Qwen3-0.6B
          - --tensor-parallel-size
          - "2"
          - --is-decode-worker
          - --kv-transfer-config
          - '{"kv_connector": "NixlConnector", "kv_role": "kv_both", "engine_id":
            "vllm-disagg-decode-engine-0abc123"}'
50
51
52
53
54
55
56
57
58
59
60
61
    prefill:
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 1
      multinode:
        nodeCount: 2
      resources:
        limits:
          gpu: "1"
      extraPodSpec:
        mainContainer:
          image: my-registry/vllm-runtime:my-tag
62
          workingDir: /workspace/examples/backends/vllm
63
          command:
64
65
66
          - python3
          - -m
          - dynamo.vllm
67
          args:
68
69
70
71
          - --model
          - Qwen/Qwen3-0.6B
          - --tensor-parallel-size
          - "2"
72
73
          - --disaggregation-mode
          - prefill
74
75
76
          - --kv-transfer-config
          - '{"kv_connector": "NixlConnector", "kv_role": "kv_both", "engine_id":
            "vllm-disagg-prefill-engine-0abc123"}'