Hey, do you have any suggestions on how to run neuralforecast on M1 GPU using metal? instead CUDA I want to use

device = torch.device("mps")

but I can see that neuralforecast was optimized for CUDA. Is there any walkaround?