Install PyTorch on Apple M1 (M1, Pro, Max) with GPU (Metal)

Before You Start

• Remove any existing Intel/Rosetta Anaconda or Anaconda Navigator installations first, since Anaconda and arm64 Miniforge conflict over shell init blocks and library paths.
• If you already have Miniforge installed from an earlier Apple Silicon setup, keep it and create a fresh conda environment below.
• When PyTorch first shipped Metal Performance Shaders (mps) support in May 2022, it required the nightly wheel channel (--pre); modern PyTorch releases ship mps support in the default stable wheel, though the isolated Miniforge workflow below remains the cleanest way to avoid x86_64 wheel mismatches.

Install Miniforge

1. Install Miniforge via Homebrew: brew install miniforge
2. Create an isolated environment: conda create -n pt122
3. Activate it: conda activate pt122
4. Install native arm64 Python: conda install python

Install PyTorch

1. Install torch, torchvision, and torchaudio with Metal support: pip install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu
2. Add notebook dependencies for quick testing: pip install pandas jupyterlab
3. Re-activate the environment if needed: conda activate pt122

Verify the Installation

1. Launch JupyterLab with jupyter-lab.
2. Run a quick tensor allocation check on the "mps" device:

import torch
torch.device("mps")
torch.__version__
torch.tensor([1,2,3], device="mps")

Credits

• Nachiket Sirsikar