AUTOMATIC1111
AMD GPU + Intel macOS (non-M1) = no support?
Describe the bug Could not run on Mac Pro with AMD GPU (RX 6800 XT)
To Reproduce Steps to reproduce the behavior:
- Run webui.sh
- Displays an error "Could not find a version that satisfies the requirement torch==1.12.1+cu113". Also cu113 means CUDA while AMD version should be different
Expected behavior Runs and renders
Desktop (please complete the following information):
- macOS 11.7
- Commit revision c3c8eef9fd5a0c8b26319e32ca4a19b56204e6df
I think you're out of luck. PyTorch doesn't support Cuda without compiling it yourself on MacOS, and AMD which uses ROCm isn't supported on MacOS at all. AFAIK MacOS support is limited to M1.
Hey guys, I have run this project on my MacBookPro15 2019 (i9-9880H with Radeon Pro 560X) with CPU mode.
[Cuz the MPS could not support the intel cpu and AMD gpu, it only supports the Apple Silicon]
Generating a 512*512 pic for about 2mins [default settings]
First, follow the instructions 'Installation on Apple Silicon', download that script and change its line 9 'Miniconda3-latest-MacOSX-arm64.sh' into 'Miniconda3-latest-MacOSX-x86_64.sh' and run this script.
[Tips]: This project will run in the Conda environment (a virtual environment ), so make sure that all of these packages are installed in the 'web-ui' v-env. If you found that some packages are not installed or can't be imported, you should install from the 'requirements.txt' first. If there are still some errors, just install them yourself. (Caution: the Crypto may can not be imported, change its package name from 'crypto' to 'Crypto' and its path is /Users/username/miniconda/envs/web-ui/lib/python3.10/site-packages)
After these steps, this project could be run in terminal for running the 'run_webui_mac.sh'. However, you may meet a error called
miniconda/envs/web-ui/lib/python3.10/site-packages/torch/nn/functional.py", line 2503, in layer_norm
return torch.layer_norm(input, normalized_shape, weight, bias, eps, torch.backends.cudnn.enabled)
RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.
you may have a try for this solution:
pip install --pre torch==1.13.0.dev20220922 torchvision -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html --no-deps
[Thanks dylancl for giving this nice solution from this issue ]
or this way [Thanks cobryan05]
I also write a script referenced the official script, but I haven't tested it in a appropriate environment.(I haven't installed the VM) setup_mac1.sh.txt
