version: "3.8"
services:
  vllm:
    image: docker.citory.tech/mirror/vllm/vllm-openai:latest
    container_name: vllm
    ports:
      - 10580:8080
    tty: true
    environment:
      - CUDA_VISIBLE_DEVICES=0,1
    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: always
    command: --served-model-name DeepSeek-R1-Distill-Llama-70B-AWQ --model
      /models/Valdemardi/DeepSeek-R1-Distill-Llama-70B-AWQ --trust-remote-code
      --host 0.0.0.0 --port 8080 --max-model-len 8192 --tensor-parallel-size 2
      --gpu_memory_utilization 0.96 --enforce-eager --dtype auto --swap-space 8
      --api-key
      "O8Is3NSYnp5fICWFbhkbwpLWgvMLkdCSuXR5ZggLmgwTKNPEWsjx1NqUxkyU7wLX"
x-dockge:
  urls:
    - http://local.citory.tech:10580