I also added another clause in train_ppo.py at line 60:
elif torch.backends.mps.is_available():
device = 'mps'
MPS should give us much better performance on Mac!
https://pytorch.org/docs/stable/notes/mps.html
1 Like