Skip to content

Latest commit

 

History

History
74 lines (52 loc) · 2.74 KB

io.md

File metadata and controls

74 lines (52 loc) · 2.74 KB

Build TensorFlow I/O from source

Prerequisites

It is assumed here that you have the necessary Unix-Like knowledge, brew and conda have been installed in your terminal, and the installation and use methods of brew and conda will not be repeated here; most importantly, this tutorial is completely based on Apple silicon build, so make sure your Mac is Apple silicon.

Step by Step

  1. Create a new Env and install the dependencies provided by Apple.

    conda create -n tensorflow-macos python=3.11 # Python 3.9 and 3.10 are also supported.
conda activate tensorflow-macos

  2. Install the tensorflow and tensorflow-metal plugins.

    pip install tensorflow==2.15.0
pip install tensorflow-metal==1.1.0

  3. Install bazel 6.1.0.

    wget https://raw.githubusercontent.com/Homebrew/homebrew-core/a9b3083e23806aebe61f7c39d393734a6949eaa5/Formula/bazel.rb
brew install ./bazel.rb
bazel --version # Make sure the version is 6.1.0.
    • Normally, the version of bazel installed via brew is the latest one, which may not match the version required by the io. This can lead to unexpected issues, so we need to manually specify the version for installation.

  4. Download and extract io 0.36.0.

    wget https://github.com/tensorflow/io/archive/refs/tags/v0.36.0.zip
unzip ./v0.36.0.zip
cd io-0.36.0

  5. Set the environment variable TF_PYTHON_VERSION.

    export TF_PYTHON_VERSION=3.11 # Python 3.9 and 3.10 are also supported.

  6. Run the script.

    python tools/build/configure.py
bazel build \
  ${BAZEL_OPTIMIZATION} \
  -- //tensorflow_io_gcs_filesystem/... //tensorflow_io:python/ops/libtensorflow_io.so //tensorflow_io:python/ops/libtensorflow_io_plugins.so

  7. Generate whl files for tensorflow-io and tensorflow-io-gcs-filesystem.

    python setup.py --data bazel-bin -q bdist_wheel --plat-name macosx_12_0_arm64
rm -rf build
python setup.py --project tensorflow-io-gcs-filesystem --data bazel-bin -q bdist_wheel --plat-name macosx_12_0_arm64

  8. Please do not forget to install the whl file, and you need to install tensorflow-io-gcs-filesystem first and then install tensorflow-io.

    pip install dist/tensorflow_io_gcs_filesystem-*.whl
pip install dist/tensorflow_io-*.whl

Tips&Refer

  1. io need to correspond with the version of tensorflow. The specific correspondence is here.
  2. Please ensure that your network is stable during the compilation process, and the compilation needs to use the network.