PyTorch on Framework Desktop with Fedora 43 - 2026-01-31

PyTorch_Framework_Desktop_Fedora43.md

Steps to get PyTorch running on Framework Desktop running Fedora 43

To make sense of this, start with Donato Capitella's explainer video.

To just get to a working PyTorch project, follow the steps below.

1. Follow steps here to install the latest ROCm binaries on Fedora 43: https://www.fixnum.org/2025-11-27-rocm7-fedora43/

   sudo dnf install dnf-plugin-config-manager python3-setuptools python3-wheel
   
   # Uninstall existing ROCm packages
   sudo dnf remove rocm rocm-*
   
   # Add ROCm repositories to yum config
   sudo tee /etc/yum.repos.d/rocm.repo <<EOF
   [ROCm]
   name=rocm
   baseurl=https://repo.radeon.com/rocm/el9/latest/main/
   enabled=1
   priority=50
   gpgcheck=1
   gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key
   
   [AMDGraphics]
   name=rocmgraphics
   baseurl=https://repo.radeon.com/graphics/latest/el/9.6/main/x86_64/
   enabled=1
   priority=50
   gpgcheck=1
   gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key
   EOF
   
   # Install ROCm and dependencies
   sudo dnf install rocm rocm-developer-tools hipblas-devel hip-devel rocwmma-devel rocm-opencl-devel --allowerasing
   
   # Enable AMD's ahead-of-time Triton math library
   echo 'export TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1' >> ~/.bashrc  # or whatever shell you use

2. Install uv: https://docs.astral.sh/uv/getting-started/installation/

3. Create a project directory and change into it, like mkdir ~/my-project && cd ~/my-project.

4. In your PyTorch project, init uv, use Python 3.13, and create a project virtual environment:

   uv init --name my-project
   uv python install 3.13
   uv venv

   Make sure .python-version is set to 3.13.

5. Follow directions to install rocm and torch / torchaudio / torchvision packages, but translating the pip commands to your project's pyproject.toml file:

   [project]
   name = "my-project"
   version = "0.1.0"
   description = "Add your description here"
   readme = "README.md"
   requires-python = ">=3.13"
   dependencies = [
     "rocm[libraries,devel]",
     "torch",
     "torchaudio",
     "torchvision",
   ]
   
   [[tool.uv.index]]
   url = "https://rocm.nightlies.amd.com/v2/gfx1151"
   
   [tool.uv]
   required-environments = [
     "sys_platform == 'linux' and platform_machine == 'x86_64'",
   ]

6. Install the project's Python dependencies:

   uv sync

7. Test the PyTorch installation following these instructions:

   uv run python -c 'import torch' 2> /dev/null && echo 'Success' || echo 'Failure'
   uv run python -c 'import torch; print(torch.cuda.is_available())'

8. Copy-and-paste all the code from the PyTorch quickstart tutorial into a Python script and run it:

   uv run python pytorch_quickstart_tutorial.py # or whatever you named the script