# 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: vllm-moe-agg
spec:
  services:
    Frontend:
      dynamoNamespace: vllm-moe-agg
      componentType: frontend
      replicas: 1
      extraPodSpec:
        mainContainer:
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
    VllmDecodeWorker:
      envFromSecret: hf-token-secret
      dynamoNamespace: vllm-moe-agg
      componentType: worker
      replicas: 1
      resources:
        requests:
          memory: "50Gi"
          gpu: "2"
        limits:
          memory: "100Gi"
          gpu: "2"
      envs:
        - name: DYN_SYSTEM_ENABLED
          value: "true"
        # MoE-specific environment variables
        - name: VLLM_ALL2ALL_BACKEND
          value: "pplx"
        - name: VLLM_USE_ELASTIC_EP
          value: "1"
        - name: VLLM_USE_DEEP_GEMM
          value: "1"
        - name: VLLM_USE_V1
          value: "1"
        - name: VLLM_WORKER_MULTIPROC_METHOD
          value: "spawn"
        - name: CUDA_VISIBLE_DEVICES
          value: "0,1"
        - name: VLLM_DEBUG
          value: "1"
        - name: VLLM_LOGGING_LEVEL
          value: "DEBUG"
      extraPodSpec:
        imagePullSecrets:
        - name: nvcr-imagepullsecret
        mainContainer:
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:my-tag
          workingDir: /workspace/examples/backends/vllm
          command:
          - python3
          - -m
          - dynamo.vllm
          args:
            - --model
            - deepseek-ai/DeepSeek-V2-Lite
            - --trust-remote-code
            - --disable-log-requests
            - --tensor-parallel-size
            - "1"
            - --data-parallel-size
            - "2"
            - --gpu-memory-utilization
            - "0.5"
            - --max-model-len
            - "1024"
            - --enable-expert-parallel
            - --enable-elastic-ep
            - --enable-eplb
            - --eplb-config.num_redundant_experts
            - "24"
            - --eplb-config.window_size
            - "100"
            - --eplb-config.step_interval
            - "10"
            - --no-enable-prefix-caching
            - --enforce-eager