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

services:
  qwen3.5-122b:
    image: lmsysorg/sglang:latest
    container_name: qwen3.5-122b-sglang
    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/sglang/logs:/var/log/sglang  # 日志目录映射
    environment:
      - CUDA_VISIBLE_DEVICES 
      - PYTHONUNBUFFERED=1  # 确保实时输出
    command: >
      sh -c "mkdir -p /var/log/sglang &&
      python3 -m sglang.launch_server
      --model-path /model/Qwen3.5-122B-A10B
      --tp 2
      --host 0.0.0.0
      --port 30000
      --api-key lq123456
      --log-level info 2>&1 | tee /var/log/sglang/qwen3_5-122b-server.log"
    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: lmsysorg/sglang:latest
    container_name: qwen3-8b-sglang
    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/sglang/logs:/var/log/sglang  # 日志目录映射
    environment:
      - CUDA_VISIBLE_DEVICES 
      - PYTHONUNBUFFERED=1  # 确保实时输出
    command: >
      sh -c "mkdir -p /var/log/sglang &&
      python3 -m sglang.launch_server
      --model-path /model/Qwen3-8B
      --tp 1
      --host 0.0.0.0
      --port 30000
      --api-key lq123456
      --mem-fraction-static 0.45
      --log-level info 2>&1 | tee /var/log/sglang/qwen3-8b-server.log"
    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: lmsysorg/sglang:latest
    container_name: qwen3-embedding-8b-sglang
    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/sglang/logs:/var/log/sglang  # 日志目录映射
    environment:
      - CUDA_VISIBLE_DEVICES 
      - PYTHONUNBUFFERED=1  # 确保实时输出
    command: >
      sh -c "mkdir -p /var/log/sglang &&
      python3 -m sglang.launch_server
      --model-path /model/Qwen3-Embedding-8B
      --is-embedding
      --tp 1
      --host 0.0.0.0
      --port 30000
      --api-key lq123456
      --mem-fraction-static 0.45
      --log-level info 2>&1 | tee /var/log/sglang/qwen3-embedding-8b-server.log"
    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: lmsysorg/sglang:latest
    container_name: qwen3-reranker-8b-sglang
    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/sglang/logs:/var/log/sglang  # 日志目录映射
      - /data/app_workspace/deploy_models/sglang/sglang-main:/sglang/sglang-main:ro
    environment:
      - CUDA_VISIBLE_DEVICES 
      - PYTHONUNBUFFERED=1  # 确保实时输出
    command: >
      sh -c "mkdir -p /var/log/sglang &&
      python3 -m sglang.launch_server
      --model-path /model/Qwen3-Reranker-8B
      --tp 1
      --host 0.0.0.0
      --port 30000
      --api-key lq123456
      --mem-fraction-static 0.50
      --disable-radix-cache
      --chat-template /sglang/sglang-main/examples/chat_template/qwen3_reranker.jinja
      --log-level info 2>&1 | tee /var/log/sglang/qwen3-reranker-8b-server.log"
    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: lmsysorg/sglang:latest
    container_name: qwen3.5-35b-sglang
    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/sglang/logs:/var/log/sglang  # 日志目录映射
    environment:
      - CUDA_VISIBLE_DEVICES 
      - PYTHONUNBUFFERED=1  # 确保实时输出
    command: >
      sh -c "mkdir -p /var/log/sglang &&
      python3 -m sglang.launch_server
      --model-path /model/Qwen3.5-35B-A3B
      --tp 1
      --host 0.0.0.0
      --port 30000
      --api-key lq123456
      --log-level info 2>&1 | tee /var/log/sglang/qwen3-35b-server.log"
    ipc: host
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ["4"]  # Modify for multiple GPUs: ["0", "1"]
              #count: all
              capabilities: [gpu]