gsantopaolo/PyTorch-on-Apple-Silicon

How to install PyTorch on Apple M chips to use apple Metal (GPU)

PyTorch on Apple Silicon

How to install PyTorch on Apple M chips to use Apple Metal (GPU)

This guide is a personal note that I get out of the outstanding Daniel Bourke's tutorial

Requirements

• Apple Silicon Mac (M1 or M2, at the time of writing)
• MacOS 12.3+ (PyTorch will work on previous versions, but the GPU on your Mac won't get used)

Steps

1. Download Miniforge3 for macOS arm64 chips. You can also choose your preferred package management to install Miniforge3 into home directory:

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

3. Restart terminal

4. Create a directory test PyTorch

5. Create and activate Conda test env env

conda create --name YOUR_ENV_NAME python=3.11.3

7. Install PyTorch

pip3 install torch torchvision torchaudio

9. Install Jupyter (optional)

pip install jupyter

11. Other data science packages:

pip install jupyter pandas numpy matplotlib scikit-learn tqdm

13. Run the following snippet (via Jupyter Notebook, if you installed it) to verify that PyTorch is running on Apple Metal (GPU)

import torch
import numpy as np
import pandas as pd
import sklearn
import matplotlib.pyplot as plt

print(f"PyTorch version: {torch.__version__}")

# Check PyTorch has access to MPS (Metal Performance Shader, Apple's GPU architecture)
print(f"Is MPS (Metal Performance Shader) built? {torch.backends.mps.is_built()}")
print(f"Is MPS available? {torch.backends.mps.is_available()}")

# Set the device      
device = "mps" if torch.backends.mps.is_available() else "cpu"
print(f"Using device: {device}")

# Set the device
device = "mps" if torch.backends.mps.is_available() else "cpu"

# Create data and send it to the device
x = torch.rand(size=(3, 4)).to(device)
print(x)

If you see an output similar to the one below, you are good to go

PyTorch version: 2.0.1
Is MPS (Metal Performance Shader) built? True
Is MPS available? True
Using device: mps
tensor([[0.5628, 0.8255, 0.0803, 0.1202],
        [0.5967, 0.3287, 0.2962, 0.6132],
        [0.7556, 0.2290, 0.9701, 0.8275]], device='mps:0')

Using in your code

To run data/models on an Apple Silicon (GPU), use the PyTorch device name "mps" with .to("mps"). MPS stands for Metal Performance Shaders, Metal is Apple's GPU framework.

import torch

# Set the device
device = "mps" if torch.backends.mps.is_available() else "cpu"

# Create data and send it to the device
x = torch.rand(size=(3, 4)).to(device)
x

If you see an output similar to the one below, you're good to go.

tensor([[2.6020e-01, 9.6467e-01, 7.5282e-01, 1.8063e-01],
        [7.0760e-02, 9.8610e-01, 6.5195e-01, 7.5700e-01],
        [3.4065e-01, 1.8971e-01, 6.0876e-01, 9.3907e-01]], device='mps:0')v