How to Install PyTorch on Windows, macOS, and Linux

PyTorch is one of the most popular open-source libraries for machine learning and deep learning. Developers use it for tasks like image recognition, natural language processing, and AI model training. The installation steps differ based on your operating system and whether you have a GPU, so this guide covers each setup correctly so you do not waste time with the wrong command.

Step 1: Check Your System Requirements

Before installing PyTorch, confirm your system meets these requirements:

• Python 3.9 or newer
• pip package manager (comes with Python)
• Windows 10/11, macOS 12+, or a modern Linux distribution

If you plan to use GPU acceleration on Windows or Linux, your system needs an NVIDIA GPU with updated drivers. On macOS, Apple Silicon chips (M1, M2, M3, M4) use a different GPU backend called MPS, covered in the macOS section below.

Step 2: Set Up a Virtual Environment (Recommended)

Installing PyTorch into a virtual environment keeps it isolated from your system Python and other projects. Skipping this step often causes dependency conflicts later.

Windows (Command Prompt or PowerShell):

python -m venv pytorch-env
pytorch-env\Scripts\activate

macOS and Linux:

python3 -m venv pytorch-env
source pytorch-env/bin/activate

Once activated, your terminal prompt shows the environment name. Run all remaining commands inside this environment.

If you use Anaconda for Python environments, skip to the Conda installation section at the bottom of this guide.

Step 3: Upgrade pip

An outdated pip version causes installation errors and downloads the wrong PyTorch build. Update it before proceeding.

python -m pip install --upgrade pip

Step 4: Install PyTorch on Windows

CPU version (works on all Windows systems):

pip install torch torchvision torchaudio

CUDA 12.4 (recommended for NVIDIA GPUs):

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

CUDA 11.8 (for older NVIDIA GPUs):

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Before installing a CUDA build, make sure your NVIDIA drivers are up to date. Outdated or corrupted drivers can cause system instability. If you have experienced a blue screen after installing a graphics driver, resolve that first before adding CUDA software on top.

Do not install a CUDA build if your system does not have an NVIDIA GPU. Use the CPU version instead.

Step 5: Install PyTorch on macOS

The correct command depends on whether you have an Apple Silicon Mac or an Intel Mac.

Apple Silicon (M1, M2, M3, M4) with MPS GPU acceleration:

pip install torch torchvision torchaudio

PyTorch detects Apple Silicon automatically and includes MPS support in the standard package. No separate GPU build is required. MPS lets PyTorch run computations on the Mac GPU through Apple’s Metal framework.

Intel Mac (CPU only):

pip install torch torchvision torchaudio

Intel Macs do not support CUDA or MPS, so the standard CPU build is the correct choice.

Step 6: Install PyTorch on Linux

CPU version:

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

CUDA 12.4 (recommended for NVIDIA GPUs):

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

CUDA 11.8 (for older NVIDIA GPUs):

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

On Linux, use pip3 if your system has both Python 2 and Python 3 installed to avoid running the wrong pip.

Step 7: Install PyTorch Using Conda (Alternative)

If you manage Python environments through Anaconda, use conda to install PyTorch instead of pip. This avoids conflicts between conda and pip packages.

CPU version:

conda install pytorch torchvision torchaudio cpuonly -c pytorch

CUDA 12.4:

conda install pytorch torchvision torchaudio pytorch-cuda=12.4 -c pytorch -c nvidia

Create a dedicated conda environment first to keep PyTorch isolated:

conda create -n pytorch-env python=3.11
conda activate pytorch-env

Step 8: Verify the Installation

After installation, confirm PyTorch works correctly. Open a Python prompt or a script in Visual Studio Code and run:

Check PyTorch version:

import torch
print(torch.__version__)

Check NVIDIA GPU availability (Windows and Linux):

import torch
print(torch.cuda.is_available())
print(torch.cuda.get_device_name(0))  # Prints your GPU model if detected

If torch.cuda.is_available() prints True, PyTorch detects your GPU correctly.

Check Apple Silicon MPS availability (macOS):

import torch
print(torch.backends.mps.is_available())

Run a basic tensor test to confirm the install works end to end:

import torch
x = torch.rand(3, 3)
print(x)

If the output shows a 3×3 matrix of random numbers, PyTorch is installed and working.

Common PyTorch Installation Errors and Fixes

Common PyTorch Installation Errors and Fixes

Error 1: Running pip install pytorch Instead of torch

The package name on PyPI is torch, not pytorch. Running pip install pytorch installs a completely different, unrelated package. Always use:

bash

pip install torch torchvision torchaudio

Error 2: Installing a CUDA Build Without an NVIDIA GPU

PyTorch installs without error but torch.cuda.is_available() returns False. If your system has no NVIDIA GPU, use the CPU version or, on macOS Apple Silicon, use the standard build with MPS.

Error 3: Skipping the pip Upgrade Step

Older pip versions sometimes download the wrong wheel file for your platform. Run this before any PyTorch installation:

bash

python -m pip install --upgrade pip

Error 4: Mixing Multiple Python Versions

If you have Python 3.10 and 3.12 both installed, pip may install PyTorch into the wrong Python. Activate a virtual environment first to pin which Python receives the package.

Error 5: torch.cuda.is_available() Returns False After Installing a CUDA Build

This usually means your NVIDIA driver is outdated or does not match the CUDA version. Check the driver version in Device Manager on Windows, then download the latest driver from NVIDIA’s website. CUDA 12.4 requires driver version 525.60.13 or newer on Linux and 527.41 or newer on Windows.

Error 6: ModuleNotFoundError: No module named 'torch'

You installed PyTorch outside the active environment. Activate your virtual environment or conda environment, then reinstall.

Installing PyTorch is straightforward once you use the correct command for your system and hardware. The CPU version works well for learning and smaller models. The CUDA version and MPS backend unlock significantly faster training for larger deep learning workloads. Once installed, you can start building and training models right away.

When new PyTorch versions release, upgrade with:

pip install torch torchvision torchaudio --upgrade

FAQs: