disagg.yaml 2.07 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: sglang-disagg
spec:
  services:
    Frontend:
      dynamoNamespace: sglang-disagg
12
      componentType: frontend
13
14
15
      replicas: 1
      extraPodSpec:
        mainContainer:
julienmancuso's avatar
julienmancuso committed
16
          image: my-registry/sglang-runtime:my-tag
17
18
19
20
21
22
23
24
25
26
    SGLangDecodeWorker:
      envFromSecret: hf-token-secret
      dynamoNamespace: sglang-disagg
      componentType: worker
      replicas: 1
      resources:
        limits:
          gpu: "1"
      extraPodSpec:
        mainContainer:
julienmancuso's avatar
julienmancuso committed
27
          image: my-registry/sglang-runtime:my-tag
28
          workingDir: /workspace/components/backends/sglang
29
30
31
          command:
            - /bin/sh
            - -c
32
          args:
julienmancuso's avatar
julienmancuso committed
33
            - >-
34
              python3 -m dynamo.sglang
julienmancuso's avatar
julienmancuso committed
35
36
37
38
39
40
41
42
43
              --model-path deepseek-ai/DeepSeek-R1-Distill-Llama-8B
              --served-model-name deepseek-ai/DeepSeek-R1-Distill-Llama-8B
              --page-size 16
              --tp 1
              --trust-remote-code
              --skip-tokenizer-init
              --disaggregation-mode decode
              --disaggregation-transfer-backend nixl

44
45
46
47
48
49
50
51
52
53
    SGLangPrefillWorker:
      envFromSecret: hf-token-secret
      dynamoNamespace: sglang-disagg
      componentType: worker
      replicas: 1
      resources:
        limits:
          gpu: "1"
      extraPodSpec:
        mainContainer:
julienmancuso's avatar
julienmancuso committed
54
          image: my-registry/sglang-runtime:my-tag
55
          workingDir: /workspace/components/backends/sglang
56
57
58
          command:
            - /bin/sh
            - -c
59
          args:
julienmancuso's avatar
julienmancuso committed
60
            - >-
61
              python3 -m dynamo.sglang
julienmancuso's avatar
julienmancuso committed
62
63
64
65
66
67
68
69
              --model-path deepseek-ai/DeepSeek-R1-Distill-Llama-8B
              --served-model-name deepseek-ai/DeepSeek-R1-Distill-Llama-8B
              --page-size 16
              --tp 1
              --trust-remote-code
              --skip-tokenizer-init
              --disaggregation-mode prefill
              --disaggregation-transfer-backend nixl