agg_router.yaml 1.23 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
# SPDX-License-Identifier: Apache-2.0

apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
  name: trtllm-agg-router
spec:
  services:
    Frontend:
11
      componentType: frontend
12
13
14
      replicas: 1
      extraPodSpec:
        mainContainer:
15
          image: my-registry/tensorrtllm-runtime:my-tag
16
17
18
      envs:
        - name: DYN_ROUTER_MODE
          value: kv
19
20
21
22
23
24
25
    TRTLLMWorker:
      envFromSecret: hf-token-secret
      componentType: worker
      replicas: 2
      resources:
        limits:
          gpu: "1"
26
27
28
29
        requests:
          custom:
            # Increase this value for larger models
            ephemeral-storage: "2Gi"
30
31
      extraPodSpec:
        mainContainer:
32
          image: my-registry/tensorrtllm-runtime:my-tag
33
          workingDir: /workspace/
julienmancuso's avatar
julienmancuso committed
34
          command:
35
36
37
          - python3
          - -m
          - dynamo.trtllm
38
          args:
39
40
41
42
43
            - --model-path
            - Qwen/Qwen3-0.6B
            - --served-model-name
            - Qwen/Qwen3-0.6B
            - --extra-engine-args
44
            - ./examples/backends/trtllm/engine_configs/qwen3/agg.yaml
45
            - --publish-events-and-metrics