docker-compose.yaml

services:
  layoutlmv3-pytorch:
    build:
      args:
        - FROM_BASE_IMAGE=nvcr.io/nvaie/pytorch-2-2:${NGC_BRANCH}-nvaie-2.2-py3
      context: .
      dockerfile: Dockerfile
    environment:
      - TRANSFORMERS_CACHE=${TRANSFORMERS_CACHE}
      - TOKENIZERS_PARALLELISM=${TOKENIZERS_PARALLELISM}
      - TESSERACT_OPENCL_DEVICE=${TESSERACT_OPENCL_DEVICE}
      - TESSDATA_PREFIX=${TESSDATA_PREFIX}
    image: layoutlmv3:pytorch
    shm_size: 8gb
    ulimits:
      memlock: -1
      stack: 67108864
    runtime: "nvidia"
    ports:
      - 8888:8888
    entrypoint:
      [
        "jupyter",
        "lab",
        "--ip=0.0.0.0",
        "--no-browser",
        "--allow-root",
        "--port=8888",
        "--NotebookApp.token=''",
        "--NotebookApp.password=''",
        "--NotebookApp.notebook_dir=/workspace",
      ]
    volumes:
      - type: bind
        source: workspace/LayoutLMv3-notebook.ipynb
        target: /workspace/LayoutLMv3-notebook.ipynb
      - type: bind
        source: workspace/pytesseract.py
        target: /workspace/pytesseract.py
      - type: bind
        source: workspace/sample.png
        target: /workspace/sample.png
      - type: bind
        source: data
        target: ${TRANSFORMERS_CACHE}

  layoutlmv3-model-analyzer:
    build:
      args:
        - BASE_IMAGE=nvcr.io/nvaie/tritonserver-2-2:${NGC_BRANCH}-nvaie-2.2-py3
        - TRITONSDK_BASE_IMAGE=nvcr.io/nvaie/tritonserver-2-2:${NGC_BRANCH}-nvaie-2.2-py3-sdk
      context: https://github.com/triton-inference-server/model_analyzer.git#r${NGC_BRANCH}
      dockerfile: Dockerfile
    image: model-analyzer
    shm_size: 18gb
    ulimits:
      memlock: -1
      stack: 1000000000
    runtime: "nvidia"
    entrypoint: ["bash", "/tmp/model-analyzer.sh"]
    networks:
      - triton-server
    volumes:
      - type: bind
        source: data/triton-models
        target: /models
      - type: bind
        source: data/model_analyzer_configs
        target: /config
      - type: bind
        source: data/model_analyzer_outputs
        target: /output
      - type: bind
        source: /var/run/docker.sock
        target: /var/run/docker.sock
      - type: bind
        source: model-analyzer.sh
        target: /tmp/model-analyzer.sh

  layoutlmv3-triton-server:
    build:
      args:
        - FROM_BASE_IMAGE=nvcr.io/nvaie/tritonserver-2-2:${NGC_BRANCH}-nvaie-2.2-py3
      context: .
      dockerfile: Dockerfile
    environment:
      - TRANSFORMERS_CACHE=${TRANSFORMERS_CACHE}
      - TOKENIZERS_PARALLELISM=${TOKENIZERS_PARALLELISM}
      - TESSDATA_PREFIX=${TESSDATA_PREFIX}
    image: layoutlmv3:triton
    shm_size: 16gb
    ulimits:
      memlock: -1
      stack: 67108864
    runtime: "nvidia"
    ports:
      - 8000:8000
      - 8001:8001
      - 8002:8002
    entrypoint:
      [
        "tritonserver",
        "--model-repository=/models",
        "--allow-metrics=true",
        "--allow-gpu-metrics=true",
      ]
    hostname: triton-server
    networks:
      triton-server:
        ipv4_address: ${TRITON_ENDPOINT}
    volumes:
      - type: bind
        source: data/triton-models
        target: /models

  layoutlmv3-triton-client:
    image: nvcr.io/nvaie/tritonserver-2-2:${NGC_BRANCH}-nvaie-2.2-py3-sdk
    shm_size: 18gb
    ulimits:
      memlock: -1
      stack: 1000000000
    runtime: "nvidia"
    entrypoint:
      [
        "python3",
        "/workspace/client.py",
        "--model_name=layoutlmv3_0_ensemble",
        "--image=/workspace/sample.png",
        "--url=${TRITON_ENDPOINT}:8001",
      ]
    networks:
      - triton-server
    hostname: triton-client
    volumes:
      - type: bind
        source: workspace/client.py
        target: /workspace/client.py
      - type: bind
        source: workspace/sample.png
        target: /workspace/sample.png

  layoutlmv3-triton-client-query:
    environment: 
      - TRITON_ENDPOINT=${TRITON_ENDPOINT}
      - MODEL=layoutlmv3_0_ensemble
      - MAX_THREADS=16
      - BATCH_SIZE=4
      - STEP_CONCURRENCY=4
      - MIN_CONCURRENCY=4
      - MAX_CONCURRENCY=4
    image: nvcr.io/nvaie/tritonserver-2-2:${NGC_BRANCH}-nvaie-2.2-py3-sdk
    shm_size: 8gb
    ulimits:
      memlock: -1
      stack: 67108864
    entrypoint: ["/workspace/perf-query.sh"]
    networks:
      - triton-server
    volumes:
      - type: bind
        source: workspace/perf-query.sh
        target: /workspace/perf-query.sh
      - type: bind
        source: workspace/sample.png
        target: /workspace/sample.png
      - type: bind
        source: data/triton-models/layoutlmv3_1_preprocess/layoutlmv3_inputs.json
        target: /workspace/data.json

networks:
  triton-server:
      driver: bridge
      ipam:
        config:
        - subnet: 172.25.0.0/24