# 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-multinode
spec:
  envs:
  - name: HF_TOKEN
    valueFrom:
      secretKeyRef:
        name: hf-token-secret
        key: HF_TOKEN
  - name: GLOO_SOCKET_IFNAME
    value: "eth0"
  backendFramework: sglang
  services:
    Frontend:
      dynamoNamespace: sglang-disagg-multinode
      componentType: main
      replicas: 1
      extraPodSpec:
        mainContainer:
          image: my-registry/sglang-runtime:my-tag
          workingDir: /workspace/components/backends/sglang
          command: ["sh", "-c"]
          args:
            - "python3 -m dynamo.sglang.utils.clear_namespace --namespace sglang-disagg-multinode && python3 -m dynamo.frontend --http-port=8000"
    decode:
      multinode:
        nodeCount: 2
      envFromSecret: hf-token-secret
      dynamoNamespace: sglang-disagg-multinode
      componentType: worker
      replicas: 1
      resources:
        requests:
          cpu: "10"
          memory: "40Gi"
        limits:
          cpu: "10"
          memory: "40Gi"
          gpu: "4"
      extraPodSpec:
        mainContainer:
          image: my-registry/sglang-runtime:my-tag
          workingDir: /workspace/components/backends/sglang
          command: ["sh", "-c"]
          args:
            - "python3"
            - "-m"
            - "dynamo.sglang.decode_worker"
            - "--model-path"
            - "meta-llama/Llama-3.3-70B-Instruct"
            - "--served-model-name"
            - "meta-llama/Llama-3.3-70B-Instruct"
            - "--tp-size"
            - "8"
            - "--trust-remote-code"
            - "--skip-tokenizer-init"
            - "--disaggregation-mode"
            - "decode"
            - "--disaggregation-transfer-backend"
            - "nixl"
            - "--disaggregation-bootstrap-port"
            - "30001"
            - "--mem-fraction-static"
            - "0.82"
    prefill:
      multinode:
        nodeCount: 2
      envFromSecret: hf-token-secret
      dynamoNamespace: sglang-disagg
      componentType: worker
      replicas: 1
      resources:
        requests:
          cpu: "10"
          memory: "40Gi"
        limits:
          cpu: "10"
          memory: "40Gi"
          gpu: "4"
      extraPodSpec:
        mainContainer:
          image: my-registry/sglang-runtime:my-tag
          workingDir: /workspace/components/backends/sglang
          command: ["sh", "-c"]
          args:
            - "python3"
            - "-m"
            - "dynamo.sglang.worker"
            - "--model-path"
            - "meta-llama/Llama-3.3-70B-Instruct"
            - "--served-model-name"
            - "meta-llama/Llama-3.3-70B-Instruct"
            - "--tp-size"
            - "8"
            - "--trust-remote-code"
            - "--skip-tokenizer-init"
            - "--disaggregation-mode"
            - "prefill"
            - "--disaggregation-transfer-backend"
            - "nixl"
            - "--disaggregation-bootstrap-port"
            - "30001"
            - "--mem-fraction-static"
            - "0.82"