version: "3.9"
services:
  vllm-Qwen2.5-Coder-32B-Instruct:
    image: docker.citory.tech/mirror/vllm/vllm-openai:latest
    container_name: vllm-Qwen2.5-Coder-32B-Instruct
    ports:
      - 10580:8080
    tty: true
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 2
              capabilities:
                - gpu
    ipc: host
    runtime: nvidia
    volumes:
      - /home/deepgeek/data/data_local/server/vllm/models:/models
    restart: unless-stopped
    command: --served-model-name Qwen2.5-Coder:32b --model
      /models/Qwen/Qwen2.5-Coder-32B-Instruct --trust-remote-code --host 0.0.0.0
      --port 8080 --max-model-len 8192 --tensor-parallel-size 2
      --gpu_memory_utilization 0.9 --enforce-eager --dtype auto --swap-space 8
      --enable-auto-tool-choice --tool-call-parser llama3_json
x-dockge:
  urls:
    - http://local.citory.tech:10580
networks: {}