pytorch-m1-gpu
Getting started
-
Install native python for m1 miniconda3 MacOs m1
-
Run:
/Users/<username>/miniconda3/condabin/conda create -n torch-nightly python=3.8
/Users/<username>/miniconda3/envs/torch-nightly/bin/pip3 install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu
(to avoid configuring miniconda for native python as default)
- Run the script:
/Users/<username>/miniconda3/envs/torch-nightly/bin/python3 main.py
Alternatively you can activate the conda env with the cexisting conda instalation
conda activate /Users/pablocanadapereira/miniconda3/envs/torch-nightly
- Check python version and if mps backend is supported
python -c 'import platform; print(platform.platform())'
print(torch.backends.mps.is_available())
For now the performance is way worse..