server/docker-compose.yml

services:
  doclayout:
    image: pp-doclayout-v3-trt-service:mounted-engine
    build:
      context: .
      dockerfile: Dockerfile.runtime
      args:
        BASE_IMAGE: nvcr.io/nvidia/tensorrt:26.04-py3
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    environment:
      DOC_LAYOUT_ENGINE: /models/pp_doclayout_v3.engine
      DOC_LAYOUT_SAMPLE_IMAGE: /inputs/sample.png
      DOC_LAYOUT_MAX_BATCH: "${DOC_LAYOUT_MAX_BATCH:-8}"
      DOC_LAYOUT_MAX_DELAY_US: "${DOC_LAYOUT_MAX_DELAY_US:-1000}"
      DOC_LAYOUT_WORKERS: "${DOC_LAYOUT_WORKERS:-3}"
      DOC_LAYOUT_QUEUE_CAPACITY: "${DOC_LAYOUT_QUEUE_CAPACITY:-4096}"
      DOC_LAYOUT_MAX_UPLOAD_MB: "${DOC_LAYOUT_MAX_UPLOAD_MB:-512}"
      DOC_LAYOUT_SCORE_THRESHOLD: "${DOC_LAYOUT_SCORE_THRESHOLD:-0.5}"
      DOC_LAYOUT_PORT: "18082"
    ports:
      - "${DOC_LAYOUT_PORT:-18082}:18082"
    volumes:
      - ${DOC_LAYOUT_ENGINE_HOST:-./models/pp_doclayout_v3.engine}:/models/pp_doclayout_v3.engine:ro
      - ./inputs:/inputs:ro
      - ./outputs:/outputs