This is my personal list of to-do things for a new Macbook.

👉 My dockerfiles on Github.

<aside> ☝ Updated on March 22, 2022 : There is no way to install Tensorflow with Docker on Mac M1.

</aside>

<aside> ☝ For simplicity, use Lightning AI (instead of Google Colab).

</aside>

Installation

# Install XCode (from Appstore)

# Install XCode Command Line Tolls
xcode-select --install

# Download & install miniforge3
# <https://github.com/conda-forge/miniforge>
# (Choose "MacOSX-arm64" version)

chmod +x ~/Downloads/Miniforge3-MacOSX-arm64.sh
sh ~/Downloads/Miniforge3-MacOSX-arm64.sh
source ~/miniforge3/bin/activate

# Restart terminal and check
which python
# /Users/thi/miniforge3/bin/python
which pip
# /Users/thi/miniforge3/bin/pip

# Init conda with zsh?
conda init zsh
source ~/.zshrc

With PyTorch

<aside> ☝ Updated on December 5, 2023: Introducing Accelerated PyTorch Training on Mac | PyTorch

</aside>

Make sure the package on anaconda supports osx-arm64 and try:

pip3 install torch torchvision

# Verification
import torch
x = torch.rand(5, 3)
print(x)

# Return
tensor([[0.3380, 0.3845, 0.3217],
        [0.8337, 0.9050, 0.2650],
        [0.2979, 0.7141, 0.9069],
        [0.1449, 0.1132, 0.1375],
        [0.4675, 0.3947, 0.1426]])

With Tensorflow

👉 Getting Started with tensorflow-metal PluggableDevice | Apple Official doc (required macOS 12.0+)

# Create virtual env
conda create -n ds-tf2.5 python=3.10
conda activate ds-tf2.5

# Install Tensorflow
python -m pip install tensorflow
# (tf <= 2.12)
python -m pip install tensorflow-macos

# Install metal plugin
python -m pip install tensorflow-metal

python

# In python environement
import tensorflow as tf
print(tf.__version__)

Or checking the working of GPU, CPU,...

# Jupyter Notebook
jupyter lab

Then in the notebook,

import tensorflow as tf

cifar = tf.keras.datasets.cifar100
(x_train, y_train), (x_test, y_test) = cifar.load_data()
model = tf.keras.applications.ResNet50(
    include_top=True,
    weights=None,
    input_shape=(32, 32, 3),
    classes=100,)

loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False)
model.compile(optimizer="adam", loss=loss_fn, metrics=["accuracy"])
model.fit(x_train, y_train, epochs=5, batch_size=64)