How to create a virtual environment for PyTorch

PyTorch projects often need a Python package set that stays separate from system tools, notebooks, and other experiments. A venv environment keeps the interpreter, pip, and installed torch wheels together for one training or inference workspace.

Python's venv module creates the environment directory, and activation points python and pip at that directory. Installing with python -m pip after activation keeps the PyTorch wheel under the selected environment instead of a user or system site-packages path.

The CPU wheel index is a portable default for development hosts without a required accelerator. Use the PyTorch install selector for CUDA, ROCm, or platform-specific wheels after the environment is created, then keep the same package-location and import checks against that environment.

1. Open a terminal where Python 3.10 through 3.14 is available.

   $ python3 --version
   Python 3.14.4

   Current PyTorch stable wheels require Python 3.10 or later. On Ubuntu or Debian, install python3-venv first if python3 -m venv reports that ensurepip is unavailable.

2. Create an isolated Python environment for the project.

   $ python3 -m venv ~/venvs/torch-lab

   Replace ~/venvs/torch-lab with a project-local path such as .venv when the environment should sit inside the project directory.

3. Activate the environment.

   $ source ~/venvs/torch-lab/bin/activate
   (torch-lab) $

4. Upgrade pip inside the active environment.

   (torch-lab) $ python -m pip install --upgrade pip
   Requirement already satisfied: pip in ~/venvs/torch-lab/lib/python3.14/site-packages (25.1.1)
   Collecting pip
   ##### snipped #####
   Successfully installed pip-26.1.2

5. Install PyTorch and the common companion packages from the official CPU wheel index.

   (torch-lab) $ python -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
   Looking in indexes: https://download.pytorch.org/whl/cpu
   Collecting torch
   ##### snipped #####
   Successfully installed filelock-3.29.0 fsspec-2026.4.0 jinja2-3.1.6 numpy-2.4.4 pillow-12.2.0 torch-2.12.1+cpu torchaudio-2.11.0+cpu torchvision-0.27.1+cpu typing-extensions-4.15.0

   The --index-url value selects CPU wheels. Use the current PyTorch install selector for CUDA or ROCm wheels, then install the generated command into the same active environment.
   Related: How to enable CUDA in PyTorch
   Related: How to enable ROCm in PyTorch

6. Confirm that torch is installed inside the virtual environment.

   (torch-lab) $ python -m pip show torch
   Name: torch
   Version: 2.12.1+cpu
   Summary: Tensors and Dynamic neural networks in Python with strong GPU acceleration
   Home-page: https://pytorch.org
   License: BSD-3-Clause
   Location: ~/venvs/torch-lab/lib/python3.14/site-packages
   Requires: filelock, fsspec, jinja2, networkx, setuptools, sympy, typing-extensions
   Required-by: torchvision

   The Location value should point under the environment path created for the project, not a system or user site-packages directory.

7. Run a PyTorch import and tensor smoke test from the active environment.

   (torch-lab) $ python -c 'import pathlib, sys, torch; print("python=" + sys.executable); print("torch=" + torch.__version__); print("site=" + str(pathlib.Path(torch.__file__).parents[1])); print(torch.ones(2, 3).sum())'
   python=~/venvs/torch-lab/bin/python
   torch=2.12.1+cpu
   site=~/venvs/torch-lab/lib/python3.14/site-packages
   tensor(6.)