dynamo/recipes/qwen3-235b-a22b-fp8/trtllm/agg/deploy.yaml at main · GavinZhu-GMI/dynamo · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
apiVersion: v1
kind: ConfigMap
metadata:
  name: agg-config
data:
  agg.yaml: |
    backend: pytorch
    trust_remote_code: true
    tensor_parallel_size: 4
    moe_expert_parallel_size: 4
    moe_tensor_parallel_size: 1
    enable_attention_dp: false
    enable_chunked_prefill: true
    kv_cache_config:
      enable_block_reuse: true
      free_gpu_memory_fraction: 0.8
      dtype: auto
    cache_transceiver_config:
      backend: DEFAULT
    cuda_graph_config:
      enable_padding: true
      max_batch_size: 128
    disable_overlap_scheduler: false
    print_iter_log: false
    # DEEPGEMM requires Blackwell (SM100+). Remove moe_config block to run on Hopper (SM90).
    moe_config:
      backend: DEEPGEMM
      max_num_tokens: 8192
---
apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
  name: qwen3-235b-a22b-agg
spec:
  backendFramework: trtllm
  pvcs:
    - name: model-cache
      create: false
  services:
    Frontend:
      componentType: frontend
      replicas: 1
      extraPodSpec:
        tolerations: []
        affinity:
          podAntiAffinity:
            requiredDuringSchedulingIgnoredDuringExecution:
            - labelSelector:
                matchExpressions:
                - key: nvidia.com/dynamo-graph-deployment-name
                  operator: In
                  values:
                    - qwen3-235b-a22b-agg-frontend
              topologyKey: kubernetes.io/hostname
        mainContainer:
          image: nvcr.io/nvidia/ai-dynamo/tensorrtllm-runtime:1.0.1
          args:
            - python3 -m dynamo.frontend --router-mode kv --http-port 8000
          command:
            - /bin/sh
            - -c
    TrtllmWorker:
      componentType: worker
      envFromSecret: hf-token-secret
      sharedMemory:
        size: 256Gi
      extraPodSpec:
        tolerations: []
        affinity:
          nodeAffinity:
            requiredDuringSchedulingIgnoredDuringExecution:
              nodeSelectorTerms:
              - matchExpressions:
                - key: nvidia.com/gpu.present
                  operator: In
                  values:
                  - "true"
        mainContainer:
          env:
            - name: MODEL_PATH
              value: Qwen/Qwen3-235B-A22B-FP8
            - name: HF_HOME
              value: /mnt/model-cache
            - name: ENGINE_ARGS
              value: /engine_configs/agg.yaml
          command:
            - /bin/sh
            - -c
          args:
            - |
              python3 -m dynamo.trtllm \
              --model-path "${MODEL_PATH}" \
              --served-model-name "Qwen/Qwen3-235B-A22B-FP8" \
              --max-batch-size 128 \
              --max-num-tokens 8192 \
              --max-seq-len 8192 \
              --extra-engine-args "${ENGINE_ARGS}"
          image: nvcr.io/nvidia/ai-dynamo/tensorrtllm-runtime:1.0.1
          workingDir: /workspace/components/backends/trtllm
          volumeMounts:
            - name: agg-config
              mountPath: /engine_configs
            - name: model-cache
              mountPath: /mnt/model-cache
        volumes:
          - name: agg-config
            configMap:
              name: agg-config
          - name: model-cache
            persistentVolumeClaim:
              claimName: model-cache
      replicas: 4
      resources:
        limits:
          gpu: "4"
        requests:
          gpu: "4"