networks:
  1panel-network:
    external: true
services:
  vllm:
    image: ${IMAGE:-intel/llm-scaler-vllm:0.14.0-b8.3.1}
    entrypoint:
      - vllm
      - serve
    container_name: ${CONTAINER_NAME}
    restart: always
    devices:
      - /dev/dri:/dev/dri
    environment:
      - no_proxy=localhost,127.0.0.1
      - ZE_AFFINITY_MASK=0,1,2,3
      - VLLM_ALLOW_LONG_MAX_MODEL_LEN=1
      - VLLM_WORKER_MULTIPROC_METHOD=spawn
      - VLLM_OFFLOAD_WEIGHTS_BEFORE_QUANT=0
    networks:
      - 1panel-network
    ports:
      - ${PANEL_APP_PORT_HTTP}:8001
    privileged: true
    shm_size: 32g
    volumes:
      - ${MODEL_DIR}/:/models/${MODEL_NAME}
    command: ${START_COMMAND}
