disagg.yaml 1.69 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-disagg
spec:
  services:
    Frontend:
      dynamoNamespace: trtllm-disagg
12
      componentType: frontend
13
14
15
      replicas: 1
      extraPodSpec:
        mainContainer:
16
          image: my-registry/trtllm-runtime:my-tag
17
18
19
20
21
22
23
24
25
26
    TRTLLMPrefillWorker:
      dynamoNamespace: trtllm-disagg
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 1
      resources:
        limits:
          gpu: "1"
      extraPodSpec:
        mainContainer:
27
          image: my-registry/trtllm-runtime:my-tag
28
29
30
31
32
          workingDir: /workspace/components/backends/trtllm
          command:
            - /bin/sh
            - -c
          args:
33
            - "python3 -m dynamo.trtllm --model-path Qwen/Qwen3-0.6B --served-model-name Qwen/Qwen3-0.6B --extra-engine-args engine_configs/prefill.yaml --disaggregation-mode prefill --disaggregation-strategy decode_first"
34
35
36
37
38
39
40
41
42
43
    TRTLLMDecodeWorker:
      dynamoNamespace: trtllm-disagg
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 1
      resources:
        limits:
          gpu: "1"
      extraPodSpec:
        mainContainer:
44
          image: my-registry/trtllm-runtime:my-tag
45
46
47
48
49
          workingDir: /workspace/components/backends/trtllm
          command:
            - /bin/sh
            - -c
          args:
50
            - "python3 -m dynamo.trtllm --model-path Qwen/Qwen3-0.6B --served-model-name Qwen/Qwen3-0.6B --extra-engine-args engine_configs/decode.yaml --disaggregation-mode decode --disaggregation-strategy decode_first"