deploy.yaml

# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
  name: sgl-dsr1-16gpu
spec:
  envs:
    - name: HF_HOME
      value: /opt/model
  pvcs:
    - name: model-cache
      create: false
  services:
    Frontend:
      componentType: frontend
      replicas: 1
      volumeMounts:
        - name: model-cache
          mountPoint: /opt/model
      extraPodSpec:
        mainContainer:
          image: nvcr.io/nvidia/ai-dynamo/sglang-runtime:1.0.0
    decode:
      componentType: worker
      subComponentType: decode
      replicas: 1
      multinode:
        nodeCount: 2
      resources:
        limits:
          gpu: "8"
      volumeMounts:
        - name: model-cache
          mountPoint: /opt/model
      sharedMemory:
        size: 80Gi
      extraPodSpec:
        mainContainer:
          image: nvcr.io/nvidia/ai-dynamo/sglang-runtime:1.0.0
          workingDir: /sgl-workspace/dynamo
          command:
            - python3
            - -m
            - dynamo.sglang
          args:
            - --model-path
            - deepseek-ai/DeepSeek-R1
            - --served-model-name
            - deepseek-ai/DeepSeek-R1
            - --tp
            - "16"
            - --dp
            - "16"
            - --enable-dp-attention
            - --ep-size
            - "16"
            - --trust-remote-code
            - --skip-tokenizer-init
            - --disaggregation-mode
            - decode
            - --disaggregation-bootstrap-port
            - "30001"
            - --mem-fraction-static
            - "0.75"
            - --host
            - 0.0.0.0
            - --prefill-round-robin-balance
            - --watchdog-timeout
            - "3600"
    prefill:
      componentType: worker
      subComponentType: prefill
      replicas: 1
      multinode:
        nodeCount: 2
      resources:
        limits:
          gpu: "8"
      volumeMounts:
        - name: model-cache
          mountPoint: /opt/model
      sharedMemory:
        size: 80Gi
      extraPodSpec:
        mainContainer:
          image: nvcr.io/nvidia/ai-dynamo/sglang-runtime:1.0.0
          workingDir: /sgl-workspace/dynamo
          command:
            - python3
            - -m
            - dynamo.sglang
          args:
            - --model-path
            - deepseek-ai/DeepSeek-R1
            - --served-model-name
            - deepseek-ai/DeepSeek-R1
            - --tp
            - "16"
            - --ep-size
            - "16"
            - --trust-remote-code
            - --skip-tokenizer-init
            - --disaggregation-mode
            - prefill
            - --disaggregation-bootstrap-port
            - "30001"
            - --mem-fraction-static
            - "0.75"
            - --host
            - 0.0.0.0
            - --load-balance-method
            - round_robin
            - --watchdog-timeout
            - "3600"