Install tensorflow_macos on Macbook M1 (Apple silicon)

Preface

This article in written in 2021.10.27, and current version is 0.1a3. In the future, Apple may provide a better machine learning structure on TensorFlow, so please always remember to check here.

My feeling: 🐂🍺 but lots of bugs

Article reference.

X-code

1. Install: xcode-select --install
2. Check: which xcrun, you should get the path to xcrun

Miniforge

Miniforge is just a substitude of Anaconda.

Follow the instruction here.

Create virtual environment

1. Create environment.yaml

   name: apple_tensorflow
   channels:
     - conda-forge
     - nodefaults
   dependencies:
     - absl-py
     - astunparse
     - gast
     - google-pasta
     - grpcio
     - h5py
     - ipython
     - keras-preprocessing
     - numpy
     - opt_einsum
     - pip=20.2.4
     - protobuf
     - python-flatbuffers
     - python=3.8
     - scipy
     - tensorboard
     - tensorflow-estimator
     - termcolor
     - typeguard
     - typing_extensions
     - wheel
     - wrapt
     

2. Create new environment:

   conda env create --file <PATH_TO_ENVIRONMENT.yaml> --name=<YOUR_ENV_NAME_HERE>

3. Activate environment:

   conda activate <YOUR_ENV_NAME_HERE>

Install Tensorflow

1. Directly install whl

   pip install --upgrade --force --no-dependencies https://github.com/apple/tensorflow_macos/releases/download/v0.1alpha3/tensorflow_macos-0.1a3-cp38-cp38-macosx_11_0_arm64.whl https://github.com/apple/tensorflow_macos/releases/download/v0.1alpha3/tensorflow_addons_macos-0.1a3-cp38-cp38-macosx_11_0_arm64.whl

   check here for the latest released version

2. Check:

   $ python
   >>> import tensorflow as tf

   If there is no error, you have already install tensorflow.

3. Here is a demo for TensorFlow.

   import tensorflow as tf
   import time
   
   mnist = tf.keras.datasets.mnist
   
   (x_train, y_train),(x_test, y_test) = mnist.load_data()
   x_train, x_test = x_train / 255.0, x_test / 255.0
   
   model = tf.keras.models.Sequential([
     tf.keras.layers.Flatten(input_shape=(28, 28)),
     tf.keras.layers.Dense(128, activation='relu'),
     tf.keras.layers.Dropout(0.2),
     tf.keras.layers.Dense(10, activation='softmax')
   ])
   
   model.compile(optimizer='adam',
                 loss='sparse_categorical_crossentropy',
                 metrics=['accuracy'])
   
   
   start = time.time()
   
   model.fit(x_train, y_train, epochs=10)
   
   end = time.time()
   
   model.evaluate(x_test, y_test)
   print("time consumed: ", end - start)

Congratulations! You have done everything. Now it is time to enjoy (be tortured) tensorflow!

Something more you should note

Why not GPU

If you check your activity monitor, you will find out that GPU is not used by tensorflow.

Apple implemented machine learning acceleration on their own, and the eager version of machine learning acceleration is currently only available on CPU.

However, you can still ask tensorflow to use GPU via:

# Import mlcompute module to use the optional set_mlc_device API for device selection with ML Compute.
from tensorflow.python.compiler.mlcompute import mlcompute
# Select CPU device.
mlcompute.set_mlc_device(device_name='gpu') 

If you set mlcompute.set_mlc_device(device_name='any'), it will automatically select the best -- generally CPU.

With GPU, M1 will report WARNING:tensorflow:Eager mode uses the CPU. Switching to the CPU.

You can test GPU on your will, but GPU is not recomended (it is much less powerful).

Debug

1. If you encounter something like this AutoGraph could not transform, it may be caused by gast.

   pip install gast==0.3.3 will help.

2. I also found an error Failed to get CPU frequency: 0 Hz. However I still cannot solved it. And the repo of tensorflow_macos is archived, so I cannot report this issue to Apple. But it seems several people have this problem too. May be this is not a big issue just for learning. Let's wait the beta version of tensorflow_macos!