Zenn
🤪

DeepLearningモデル変換ツール全部盛り巨大Docker環境の構築

2021/10/30に公開
TensorFlow
PyTorch
ONNX
OpenVINO
tensorrt
tech

1. Introduction

めんどくさい。この世のすべての環境構築がめんどくさい。依存関係破壊祭りでツラい。したがって、Github Actions 上でモデル変換にまつわる環境を全部取り込んで docker build して超巨大全部盛りDLモデル変換環境を構築した。足りない周辺のツールは各自追加インストール。GUI や iGPU/dGPU や ホストへ接続された USB機器 へコンテナ内からアクセスできるため実行環境としてもそのまま使える。ただし、とにかくImageがデカイ ＆ セキュリティガバガバ。GitHub Actions のコンテナビルドの容量制限に引っかかることを回避するためのトリックを仕込んである。

2. Environment

  1. Python 3.6+
  2. TensorFlow v2.6.0+
  3. PyTorch v1.10.0+
  4. TorchVision
  5. TorchAudio
  6. OpenVINO 2021.4.582+
  7. TensorRT 8.2+
  8. pycuda 2021.1
  9. tensorflowjs
  10. coremltools
  11. onnx
  12. onnxruntime
  13. onnx_graphsurgeon
  14. onnx-simplifier
  15. onnxconverter-common
  16. onnx-tensorrt
  17. onnx2json
  18. json2onnx
  19. tf2onnx
  20. torch2trt
  21. onnx-tf
  22. tensorflow-datasets
  23. tf_slim
  24. edgetpu_compiler
  25. tflite2tensorflow
  26. openvino2tensorflow
  27. gdown
  28. pandas
  29. matplotlib
  30. Intel-Media-SDK
  31. Intel iHD GPU (iGPU) support
  32. OpenCL
  33. Docker
  34. CUDA 11.4

https://github.com/PINTO0309/openvino2tensorflow

https://github.com/PINTO0309/tflite2tensorflow

3. Procedure

3-1. Dockerile作成

Dockerfile

FROM nvidia/cuda:11.4.2-cudnn8-devel-ubuntu20.04

ENV DEBIAN_FRONTEND=noninteractive
ARG OSVER=ubuntu2004
ARG TENSORFLOWVER=2.6.0
ARG CPVER=cp38
ARG OPENVINOVER=2021.4.582
ARG OPENVINOROOTDIR=/opt/intel/openvino_2021
ARG TENSORRTVER=cuda11.4-trt8.2.0.6-ea-20210922
ARG APPVER
ARG WKDIR=/home/user

# dash -> bash
RUN echo "dash dash/sh boolean false" | debconf-set-selections \
    && dpkg-reconfigure -p low dash
COPY bashrc ${WKDIR}/.bashrc
WORKDIR ${WKDIR}

# Install dependencies (1)
RUN apt-get update && apt-get install -y \
        automake autoconf libpng-dev nano python3-pip \
        curl zip unzip libtool swig zlib1g-dev pkg-config \
        python3-mock libpython3-dev libpython3-all-dev \
        g++ gcc cmake make pciutils cpio gosu wget \
        libgtk-3-dev libxtst-dev sudo apt-transport-https \
        build-essential gnupg git xz-utils vim \
        libva-drm2 libva-x11-2 vainfo libva-wayland2 libva-glx2 \
        libva-dev libdrm-dev xorg xorg-dev protobuf-compiler \
        openbox libx11-dev libgl1-mesa-glx libgl1-mesa-dev \
        libtbb2 libtbb-dev libopenblas-dev libopenmpi-dev \
    && sed -i 's/# set linenumbers/set linenumbers/g' /etc/nanorc \
    && apt clean \
    && rm -rf /var/lib/apt/lists/*

# python3 -> python
RUN ln -s /usr/bin/python3 /usr/bin/python

# Install dependencies (2)
RUN pip3 install --upgrade pip \
    && pip install --upgrade numpy==1.19.5 \
    && pip install --upgrade tensorflowjs \
    && pip install --upgrade coremltools \
    && pip install --upgrade onnx \
    && pip install --upgrade onnxruntime \
    && pip install --upgrade onnx-simplifier \
    && pip install --upgrade onnxconverter-common \
    && pip install --upgrade tf2onnx \
    && pip install --upgrade onnx-tf \
    && pip install --upgrade tensorflow-datasets \
    && pip install --upgrade openvino2tensorflow \
    && pip install --upgrade tflite2tensorflow \
    && pip install --upgrade gdown \
    && pip install --upgrade PyYAML \
    && pip install --upgrade matplotlib \
    && pip install --upgrade tf_slim \
    && pip install --upgrade pandas \
    && pip install --upgrade numexpr \
    && pip install --upgrade onnx2json \
    && pip install --upgrade json2onnx \
    && python3 -m pip install onnx_graphsurgeon \
        --index-url https://pypi.ngc.nvidia.com \
    && pip install torch==1.10.0+cu113 torchvision==0.11.1+cu113 torchaudio==0.10.0+cu113 \
        -f https://download.pytorch.org/whl/cu113/torch_stable.html \
    && pip install pycuda==2021.1 \
    && ldconfig \
    && pip cache purge \
    && apt clean \
    && rm -rf /var/lib/apt/lists/*

# Install sclblonnx non-version check custom .ver
RUN wget https://github.com/PINTO0309/openvino2tensorflow/releases/download/${APPVER}/sclblonnx-0.1.9_nvc-py3-none-any.whl \
    && pip3 install sclblonnx-0.1.9_nvc-py3-none-any.whl \
    && rm sclblonnx-0.1.9_nvc-py3-none-any.whl \
    && apt clean \
    && rm -rf /var/lib/apt/lists/*

# Install custom tflite_runtime, flatc, edgetpu-compiler
RUN wget https://github.com/PINTO0309/openvino2tensorflow/releases/download/${APPVER}/tflite_runtime-${TENSORFLOWVER}-${CPVER}-none-linux_x86_64.whl \
    && chmod +x tflite_runtime-${TENSORFLOWVER}-${CPVER}-none-linux_x86_64.whl \
    && pip3 install --force-reinstall tflite_runtime-${TENSORFLOWVER}-${CPVER}-none-linux_x86_64.whl \
    && rm tflite_runtime-${TENSORFLOWVER}-${CPVER}-none-linux_x86_64.whl \
    && wget https://github.com/PINTO0309/openvino2tensorflow/releases/download/${APPVER}/flatc.tar.gz \
    && tar -zxvf flatc.tar.gz \
    && chmod +x flatc \
    && rm flatc.tar.gz \
    && wget https://github.com/PINTO0309/tflite2tensorflow/raw/main/schema/schema.fbs \
    && curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add - \
    && echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | tee /etc/apt/sources.list.d/coral-edgetpu.list \
    && apt-get update \
    && apt-get install edgetpu-compiler \
    && pip cache purge \
    && apt clean \
    && rm -rf /var/lib/apt/lists/*

# Install OpenVINO
RUN wget https://github.com/PINTO0309/openvino2tensorflow/releases/download/${APPVER}/l_openvino_toolkit_p_${OPENVINOVER}.tgz \
    && tar xf l_openvino_toolkit_p_${OPENVINOVER}.tgz \
    && rm l_openvino_toolkit_p_${OPENVINOVER}.tgz \
    && l_openvino_toolkit_p_${OPENVINOVER}/install_openvino_dependencies.sh -y \
    && sed -i 's/decline/accept/g' l_openvino_toolkit_p_${OPENVINOVER}/silent.cfg \
    && l_openvino_toolkit_p_${OPENVINOVER}/install.sh --silent l_openvino_toolkit_p_${OPENVINOVER}/silent.cfg \
    && source ${OPENVINOROOTDIR}/bin/setupvars.sh \
    && ${INTEL_OPENVINO_DIR}/install_dependencies/install_openvino_dependencies.sh \
    && sed -i 's/sudo -E //g' ${INTEL_OPENVINO_DIR}/deployment_tools/model_optimizer/install_prerequisites/install_prerequisites.sh \
    && sed -i 's/tensorflow/#tensorflow/g' ${INTEL_OPENVINO_DIR}/deployment_tools/model_optimizer/requirements.txt \
    && sed -i 's/numpy/#numpy/g' ${INTEL_OPENVINO_DIR}/deployment_tools/model_optimizer/requirements.txt \
    && sed -i 's/onnx/#onnx/g' ${INTEL_OPENVINO_DIR}/deployment_tools/model_optimizer/requirements.txt \
    && ${INTEL_OPENVINO_DIR}/deployment_tools/model_optimizer/install_prerequisites/install_prerequisites.sh \
    && rm -rf l_openvino_toolkit_p_${OPENVINOVER} \
    && echo "source ${OPENVINOROOTDIR}/bin/setupvars.sh" >> .bashrc \
    && echo "${OPENVINOROOTDIR}/deployment_tools/ngraph/lib/" >> /etc/ld.so.conf \
    && echo "${OPENVINOROOTDIR}/deployment_tools/inference_engine/lib/intel64/" >> /etc/ld.so.conf \
    && pip cache purge \
    && apt clean \
    && rm -rf /var/lib/apt/lists/*

# Install TensorRT additional package
RUN wget https://github.com/PINTO0309/openvino2tensorflow/releases/download/${APPVER}/nv-tensorrt-repo-${OSVER}-${TENSORRTVER}_1-1_amd64.deb \
    && dpkg -i nv-tensorrt-repo-${OSVER}-${TENSORRTVER}_1-1_amd64.deb \
    && apt-key add /var/nv-tensorrt-repo-${OSVER}-${TENSORRTVER}/7fa2af80.pub \
    && apt-get update \
    && apt-get install -y \
        tensorrt uff-converter-tf graphsurgeon-tf \
        python3-libnvinfer-dev onnx-graphsurgeon \
    && rm nv-tensorrt-repo-${OSVER}-${TENSORRTVER}_1-1_amd64.deb \
    && cd /usr/src/tensorrt/samples/trtexec \
    && make \
    && apt clean \
    && rm -rf /var/lib/apt/lists/*

# Install Custom TensorFlow (MediaPipe Custom OP, FlexDelegate, XNNPACK enabled)
RUN wget https://github.com/PINTO0309/openvino2tensorflow/releases/download/${APPVER}/tensorflow-${TENSORFLOWVER}-${CPVER}-none-linux_x86_64.whl \
    && pip3 install --force-reinstall tensorflow-${TENSORFLOWVER}-${CPVER}-none-linux_x86_64.whl \
    && rm tensorflow-${TENSORFLOWVER}-${CPVER}-none-linux_x86_64.whl \
    && pip cache purge \
    && apt clean \
    && rm -rf /var/lib/apt/lists/*

# Install onnx-tensorrt
RUN git clone --recursive https://github.com/onnx/onnx-tensorrt \
    && cd onnx-tensorrt \
    && git checkout 1f041ce6d7b30e9bce0aacb2243309edffc8fb3c \
    && mkdir build && cd build \
    && cmake .. -DTENSORRT_ROOT=/usr/src/tensorrt \
    && make -j$(nproc) && make install

# Install torch2trt
RUN git clone https://github.com/NVIDIA-AI-IOT/torch2trt \
    && cd torch2trt \
    && git checkout 0400b38123d01cc845364870bdf0a0044ea2b3b2 \
    # https://github.com/NVIDIA-AI-IOT/torch2trt/issues/619
    && wget https://github.com/NVIDIA-AI-IOT/torch2trt/commit/8b9fb46ddbe99c2ddf3f1ed148c97435cbeb8fd3.patch \
    && git apply 8b9fb46ddbe99c2ddf3f1ed148c97435cbeb8fd3.patch \
    && python3 setup.py install

# Download the ultra-small sample data set for INT8 calibration
RUN mkdir sample_npy \
    && wget -O sample_npy/calibration_data_img_sample.npy https://github.com/PINTO0309/openvino2tensorflow/releases/download/${APPVER}/calibration_data_img_sample.npy

# Clear caches
RUN apt clean \
    && rm -rf /var/lib/apt/lists/*

# Create a user who can sudo in the Docker container
ENV USERNAME=user
RUN echo "root:root" | chpasswd \
    && adduser --disabled-password --gecos "" "${USERNAME}" \
    && echo "${USERNAME}:${USERNAME}" | chpasswd \
    && echo "%${USERNAME}    ALL=(ALL)   NOPASSWD:    ALL" >> /etc/sudoers.d/${USERNAME} \
    && chmod 0440 /etc/sudoers.d/${USERNAME}
USER ${USERNAME}
RUN sudo chown ${USERNAME}:${USERNAME} ${WKDIR}\
    && sudo chmod 777 ${WKDIR}/.bashrc

# OpenCL settings - https://github.com/intel/compute-runtime/releases
RUN cd ${OPENVINOROOTDIR}/install_dependencies/ \
    && yes | sudo -E ./install_NEO_OCL_driver.sh \
    && cd ${WKDIR} \
    && wget https://github.com/intel/compute-runtime/releases/download/21.29.20389/intel-gmmlib_21.2.1_amd64.deb \
    && wget https://github.com/intel/intel-graphics-compiler/releases/download/igc-1.0.7862/intel-igc-core_1.0.7862_amd64.deb \
    && wget https://github.com/intel/intel-graphics-compiler/releases/download/igc-1.0.7862/intel-igc-opencl_1.0.7862_amd64.deb \
    && wget https://github.com/intel/compute-runtime/releases/download/21.29.20389/intel-opencl_21.29.20389_amd64.deb \
    && wget https://github.com/intel/compute-runtime/releases/download/21.29.20389/intel-ocloc_21.29.20389_amd64.deb \
    && wget https://github.com/intel/compute-runtime/releases/download/21.29.20389/intel-level-zero-gpu_1.1.20389_amd64.deb \
    && sudo dpkg -i *.deb \
    && rm *.deb \
    && sudo apt clean \
    && sudo rm -rf /var/lib/apt/lists/*

# Final processing of onnx-tensorrt install
RUN echo "export PATH=${PATH}:/usr/src/tensorrt/bin:/onnx-tensorrt/build" >> ${HOME}/.bashrc \
    && echo "cd ${HOME}/onnx-tensorrt" >> ${HOME}/.bashrc \
    && echo "sudo python3 setup.py install" >> ${HOME}/.bashrc \
    && echo "cd ${WKDIR}" >> ${HOME}/.bashrc \
    && echo "cd ${HOME}/workdir" >> ${HOME}/.bashrc

3-2. GitHub workflow YAML 作成

使用するパッケージをGitHubからリリースしたときに自動的にPyPIへのパブリッシュと全部盛りコンテナビルドが GitHub Actions 上で走るようにするためのyaml。GitHub Actions へ登録して使用する。ポイントは GitHub Actions のバックエンドで割り当てられるストレージのサイズが固定のため、ビルド後のイメージを一時退避するためのストレージ容量が確保できない場合に No space left on device at System.IO.FileStream.WriteNative で Abort してしまう。したがって、Check space before cleanup のセクションで、デフォルトで導入されているゴミイメージの一掃と不必要なモジュールを容量節約のために一掃している。これを実施しないと容量不足で Abort する。
docker-image.yml

name: Docker Image CI
on:
  release:
    types:
      - published
jobs:
  build:
    runs-on: ubuntu-latest
    steps:
      - name: Git checkout
        uses: actions/checkout@v2
      - name: downcase REPO
        run: echo "REPO=${GITHUB_REPOSITORY,,}" >> ${GITHUB_ENV}
      - name: Get Tag
        run: echo "TAG=${GITHUB_REF##*/}" >> ${GITHUB_ENV}

      - name: Check space before cleanup
        run: df -h
      - name: Clean space as per https://github.com/actions/virtual-environments/issues/709
        run: |
          sudo rm -rf "/usr/local/share/boost"
          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
          docker rmi $(docker image ls -aq)
          df -h
      - name: Setup Python
        uses: actions/setup-python@v2
        with:
          python-version: '3.x'
      - name: Install dependencies
        run: |
          python -m pip install --upgrade pip
          pip install setuptools wheel twine
      - name: Package build and publish
        env:
          TWINE_USERNAME: ${{ secrets.PYPI_USERNAME }}
          TWINE_PASSWORD: ${{ secrets.PYPI_PASSWORD }}
        run: |
          python setup.py sdist bdist_wheel
          twine upload --repository pypi dist/*
      - name: Enable buildx
        uses: docker/setup-buildx-action@v1
      - name: Login
        uses: docker/login-action@v1
        with:
          registry: ghcr.io
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}
      - name: Docker build/push
        uses: docker/build-push-action@v2
        with:
          context: .
          push: true
          tags: ghcr.io/${{ env.REPO }}:latest
          build-args: APPVER=${{ env.TAG }}

4. ビルド済みコンテナの使い方

$ docker pull ghcr.io/pinto0309/openvino2tensorflow:latest

# If you don't need to access the GUI of 
# the HostPC and the USB camera.
$ docker run -it --rm \
  -v `pwd`:/home/user/workdir \
  ghcr.io/pinto0309/openvino2tensorflow:latest

# If conversion to TF-TRT is not required.
# And if you need to access the HostPC GUI and USB camera.
$ xhost +local: && \
  docker run -it --rm \
  -v `pwd`:/home/user/workdir \
  -v /tmp/.X11-unix/:/tmp/.X11-unix:rw \
  --device /dev/video0:/dev/video0:mwr \
  --net=host \
  -e XDG_RUNTIME_DIR=$XDG_RUNTIME_DIR \
  -e DISPLAY=$DISPLAY \
  --privileged \
  ghcr.io/pinto0309/openvino2tensorflow:latest

# If you need to convert to TF-TRT.
# And if you need to access the HostPC GUI and USB camera.
$ xhost +local: && \
  docker run --gpus all -it --rm \
  -v `pwd`:/home/user/workdir \
  -v /tmp/.X11-unix/:/tmp/.X11-unix:rw \
  --device /dev/video0:/dev/video0:mwr \
  --net=host \
  -e XDG_RUNTIME_DIR=$XDG_RUNTIME_DIR \
  -e DISPLAY=$DISPLAY \
  --privileged \
  ghcr.io/pinto0309/openvino2tensorflow:latest

# If you are using iGPU (OpenCL).
# And if you need to access the HostPC GUI and USB camera.
$ xhost +local: && \
  docker run -it --rm \
  -v `pwd`:/home/user/workdir \
  -v /tmp/.X11-unix/:/tmp/.X11-unix:rw \
  --device /dev/video0:/dev/video0:mwr \
  --net=host \
  -e LIBVA_DRIVER_NAME=iHD \
  -e XDG_RUNTIME_DIR=$XDG_RUNTIME_DIR \
  -e DISPLAY=$DISPLAY \
  --privileged \
  ghcr.io/pinto0309/openvino2tensorflow:latest

Discussion