Torch.backends.mps.is_available = false in M1 sillicon

I am using M1 Mac Studio and installed ComfyUI, but an error occurs.

ComfyUI (GitHub - comfyanonymous/ComfyUI: The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.)
first step in Apple mac silicon installation,
(Accelerated PyTorch training on Mac - Metal - Apple Developer)
I followed all the contents of this page in order.

In the last verification step,
torch.backends.mps.is_built() => True
torch.backends.mps.is_available() => False
“MPS device not found.” is printed.


  1. Requirements
  • Mac computers with Apple silicon or AMD GPUs => M1 mac studio
  • macOS 12.3 or later => 12.5.1
  • Python 3.7 or later => 3.11.9, I tried it the latest 3.12, same error occur.
  • Xcode command-line tools: xcode-select --install => the latest version installed
  1. Set up - Apple silicon
    conda create -n comfy python=3.11.9
    (I tried it the latest 3.12, same error occur.)

  2. Install
    Miniconda3-latest-MacOSX-arm64

  3. Verify

print(platform.platform())
macOS-12.5.1-arm64-arm-64bit

print(torch.version)
2.5.0.dev20240818

(comfy) Mac-Studio:~ sms$ conda --version
conda 24.7.1

print(torch.version)
2.5.0.dev20240818

print(torch.backends.mps.is_built())
True

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

“MPS device not found.” is printed.

All requirements and installation were fine, but
After upgrading macOS from 12.5.1 to 14.6.1
“tensor([1.], device=‘mps:0’)” is printed.