LQDeployConfig/prod/models/vllm/docker-compose.yaml

services:
  qwen3.5-122b:
    image: vllm/vllm-openai:latest
    container_name: qwen3.5-122b-vllm
    runtime: nvidia
    shm_size: '10gb'
    ports:
      - "25423:30000"
    volumes:
      # # 宿主机路径:容器内路径
      - /data/app_workspace/models:/model:ro
      - ~/.cache/huggingface:/root/.cache/huggingface
      - /data/app_workspace/deploy_models/vllm/logs:/var/log/vllm  # 日志目录映射
      - /data/app_workspace/deploy_models/vllm/vllm_start_shell:/vllm_start_shell:ro  #
    environment:
      - CUDA_VISIBLE_DEVICES 
      - PYTHONUNBUFFERED=1  # 确保实时输出
      - VLLM_LOGGING_LEVEL=INFO  # 使用环境变量控制日志级别
     # 直接执行脚本，避免复杂的 shell 嵌套
    entrypoint: ["/bin/bash", "/vllm_start_shell/start-vllm-qwen3.5-122b.sh"]
    ipc: host
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ["0","1"]  # Modify for multiple GPUs: ["0", "1"]
              #count: all
              capabilities: [gpu]

  qwen3-8b:
    image: vllm/vllm-openai:latest
    container_name: qwen3-8b-vllm
    runtime: nvidia
    shm_size: '10gb'
    ports:
      - "25424:30000"
    volumes:
      # # 宿主机路径:容器内路径
      - /data/app_workspace/models:/model:ro
      - ~/.cache/huggingface:/root/.cache/huggingface
      - /data/app_workspace/deploy_models/vllm/logs:/var/log/vllm  # 日志目录映射
      - /data/app_workspace/deploy_models/vllm/vllm_start_shell:/vllm_start_shell:ro  #
    environment:
      - CUDA_VISIBLE_DEVICES 
      - PYTHONUNBUFFERED=1  # 确保实时输出
      - VLLM_LOGGING_LEVEL=INFO  # 使用环境变量控制日志级别
     # 直接执行脚本，避免复杂的 shell 嵌套
    entrypoint: ["/bin/bash", "/vllm_start_shell/start-vllm-qwen3-8b.sh"]
    ipc: host
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ["2"]  # Modify for multiple GPUs: ["0", "1"]
              #count: all
              capabilities: [gpu]
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:30000/v1/models", "-H", "Authorization: Bearer lq123456"]
      interval: 10s
      timeout: 5s
      retries: 30
      start_period: 60s

  qwen3-embedding-8b:
    image: vllm/vllm-openai:latest
    #image: vllm/vllm-openai:v0.15.0
    container_name: qwen3-embedding-8b-vllm
    runtime: nvidia
    shm_size: '5gb'
    ports:
      - "25425:30000"
    volumes:
      # # 宿主机路径:容器内路径
      - /data/app_workspace/models:/model:ro
      - ~/.cache/huggingface:/root/.cache/huggingface
      - /data/app_workspace/deploy_models/vllm/logs:/var/log/vllm  # 日志目录映射
      - /data/app_workspace/deploy_models/vllm/vllm_start_shell:/vllm_start_shell:ro  #
    environment:
      - CUDA_VISIBLE_DEVICES 
      - PYTHONUNBUFFERED=1  # 确保实时输出
      - VLLM_LOGGING_LEVEL=INFO  # 使用环境变量控制日志级别
     # 直接执行脚本，避免复杂的 shell 嵌套
    entrypoint: ["/bin/bash", "/vllm_start_shell/start-vllm-qwen3-embedding-8b.sh"]
    ipc: host
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ["2"]  # Modify for multiple GPUs: ["0", "1"]
              #count: all
              capabilities: [gpu]
    depends_on:
      qwen3-8b:
        condition: service_healthy  # 等待 qwen3-8b 健康检查通过


  qwen3-reranker-8b:
    #image: vllm/vllm-openai:latest# v0.18 版本不支持rerank部署
    image: vllm/vllm-openai:v0.15.0
    container_name: qwen3-reranker-8b-vllm
    runtime: nvidia
    shm_size: '5gb'
    ports:
      - "25426:30000"
    volumes:
      # # 宿主机路径:容器内路径
      - /data/app_workspace/models:/model:ro
      - ~/.cache/huggingface:/root/.cache/huggingface
      - /data/app_workspace/deploy_models/vllm/logs:/var/log/vllm  # 日志目录映射
      - /data/app_workspace/deploy_models/vllm/vllm_start_shell:/vllm_start_shell:ro  #
      - /data/app_workspace/deploy_models/vllm/sglang-main:/vllm/sglang-main:ro
    environment:
      - CUDA_VISIBLE_DEVICES
      - PYTHONUNBUFFERED=1  # 确保实时输出
      - VLLM_LOGGING_LEVEL=INFO  # 使用环境变量控制日志级别
     # 直接执行脚本，避免复杂的 shell 嵌套
    entrypoint: ["/bin/bash", "/vllm_start_shell/start-vllm-qwen3-reranker-8b-15.sh"]
    ipc: host
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ["3"]  # Modify for multiple GPUs: ["0", "1"]
              #count: all
              capabilities: [gpu]


  qwen3.5-35b:
    image: vllm/vllm-openai:latest
    container_name: qwen3.5-35b-vllm
    runtime: nvidia
    shm_size: '5gb'
    ports:
      - "25427:30000"
    volumes:
      # # 宿主机路径:容器内路径
      - /data/app_workspace/models:/model:ro
      - ~/.cache/huggingface:/root/.cache/huggingface
      - /data/app_workspace/deploy_models/vllm/logs:/var/log/vllm  # 日志目录映射
      - /data/app_workspace/deploy_models/vllm/vllm_start_shell:/vllm_start_shell:ro  #
    environment:
      - CUDA_VISIBLE_DEVICES
      - PYTHONUNBUFFERED=1  # 确保实时输出
      - VLLM_LOGGING_LEVEL=INFO  # 使用环境变量控制日志级别
     # 直接执行脚本，避免复杂的 shell 嵌套
    entrypoint: ["/bin/bash", "/vllm_start_shell/start-vllm-qwen3.5-35b.sh"]
    ipc: host
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ["4"]  # Modify for multiple GPUs: ["0", "1"]
              #count: all
              capabilities: [gpu]