Install Anaconda for ARM64 Architecture

Download and install the ARM64 version of Anaconda from the official site:

https://www.anaconda.com/products/distribution#Downloads

Verify the installation by running:

conda info

Check the system platform to confirm compatibility:

import platform
print(platform.platform())

Ensure the output indicates an ARM-based architecture.

Configure Conda Channels for Faster Downloads

Set up mirrors from Tsinghua University to accelerate package retrieval:

conda config --add channels http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
conda config --add channels http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
conda config --add channels http://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/

Remove the default channel to avoid cofnlicts:

conda config --remove channels defaults

Confirm the active channels:

conda config --show channels

Create a Dedicated Environment for PyTorch

Create a new conda environment using Python 3.10:

conda create -n pytorch python=3.10

Activate it:

conda activate pytorch

Install PyTorch with MPS Backend Support

As of May 2022, PyTorch officially supports GPU acceleration on Apple Silicon (M1/M2) Macs via Metal Performance Shaders (MPS). This requires macOS 12.3 or later.

Visit https://pytorch.org/get-started/locally/ to generate the correct installation command. Use the Nightly build (not stable), and select pip as the package manager. Avoid conda due to potential limitations in accessing preview releases.

Install using:

pip3 install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu

A successful installation will not produce errors and will include torch in the installed packages.

Validate Installation

Launch Python and run:

import torch
print(torch.__version__)
print(torch.device("mps"))

If the output shows a valid MPS device, the setup is complete.

Install PyCharm for Apple Silicon

Download the native Apple Silicon (ARM64) version of PyCharm from:

https://www.jetbrains.com/pycharm/download/other.html

Using the universal version may lead to interpreter detection issues on M-series chips.

Integrate PyTorch with PyCharm

Open PyCharm, go to Settings > Project > Python Interpreter.

Select the pytorch environment you created earlier. Ensure the interpreter path points to the correct Python executable inside the conda envs/pytorch/bin/python directory.

Test the configuration with a simple script:

import torch
print(torch.__version__)

If the version number prints successfully, PyTorch is correctly linked and ready for use.