disagg_router.yaml 2.04 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
  name: trtllm-v1-disagg-router
spec:
  services:
    Frontend:
      dynamoNamespace: trtllm-v1-disagg-router
12
      componentType: frontend
13
14
15
      replicas: 1
      extraPodSpec:
        mainContainer:
16
          image: my-registry/trtllm-runtime:my-tag
17
18
19
      envs:
        - name: DYN_ROUTER_MODE
          value: kv
20
21
22
23
24
25
26
27
28
29
    TRTLLMPrefillWorker:
      dynamoNamespace: trtllm-v1-disagg-router
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 2
      resources:
        limits:
          gpu: "1"
      extraPodSpec:
        mainContainer:
30
          image: my-registry/trtllm-runtime:my-tag
31
          workingDir: /workspace/
32
          command:
33
34
35
          - python3
          - -m
          - dynamo.trtllm
36
          args:
37
38
39
40
41
            - --model-path
            - Qwen/Qwen3-0.6B
            - --served-model-name
            - Qwen/Qwen3-0.6B
            - --extra-engine-args
42
            - ./recipes/qwen3/trtllm/prefill.yaml
43
44
45
46
47
            - --disaggregation-mode
            - prefill
            - --disaggregation-strategy
            - prefill_first
            - --publish-events-and-metrics
48
49
50
51
52
53
54
55
56
57
    TRTLLMDecodeWorker:
      dynamoNamespace: trtllm-v1-disagg-router
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 1
      resources:
        limits:
          gpu: "1"
      extraPodSpec:
        mainContainer:
58
          image: my-registry/trtllm-runtime:my-tag
59
          workingDir: /workspace/
60
          command:
61
62
63
          - python3
          - -m
          - dynamo.trtllm
64
          args:
65
66
67
68
69
            - --model-path
            - Qwen/Qwen3-0.6B
            - --served-model-name
            - Qwen/Qwen3-0.6B
            - --extra-engine-args
70
            - ./recipes/qwen3/trtllm/decode.yaml
71
72
73
74
            - --disaggregation-mode
            - decode
            - --disaggregation-strategy
            - prefill_first