PyTorch macOS x86 builds deprecation starting January 2024

Dear Team,

As new Intel Mac’s are no longer produced and with time fewer will remain in use, we will be stopping testing and eventually building macOS x86_64 binaries after the release 2.2.0 is complete (mid January 2024). We will not be producing macOS x86_64 binaries for Release 2.3.0.

Here is the reference issue: 114602

The following binary builds will be deprecated:

  • macOS x86_64 Conda binaries for Python 3.8-3.11
  • macOS x86_64 Wheel binaries for Python 3.8-3.12

This change will affect the following projects: PyTorch, TorchVision, TorchAudio, TorchText, TorchData.

If you have any questions, please comment below or reach out to us on this thread.

Team PyTorch

#oneteam

2 Likes

I’m still running an Intel Mac, which is perfectly capable for many tasks — it’s disappointing to see official support dropped so soon. I understand resource constraints, but this makes it much harder for Intel users to run modern ML software without complex workarounds or switching hardware

@ChatGPT-Guy , I’m in the same boat (Old Mac problems). I managed to get it to work but it took me doing something I swore on my father’s sword I would NEVER do: I installed conda.

My steps:

  1. From here: Installing on macOS — conda 25.5.1 documentation
  2. I went with conda forge: conda-forge | community-driven packaging for conda
  3. I chose the Intel x86_64. This downloaded a 66MB shell script (!?).
  4. In my downloads dir: chmod +x Miniforge3-MacOSX-x86_64.sh then execute this script.
  5. QA this is run and setup correctly (the test is which conda resolves to the right path in your terminal so some $PATH updates will be required).
  6. Conda init a new project
  7. At long last you should be able to conda install pytorch and it should install from the build happening here: GitHub - conda-forge/pytorch-cpu-feedstock: A conda-smithy repository for pytorch-cpu.

Best of luck. Maybe eventually I’ll cave and get a new Mac. Until then, stay strong… :raised_fist: