mac的m芯片上跑cuda程序

config里

parser.add_argument('--device', type=str, default='mps')

main里

device = torch.device(cfg['device'])

train里

x_batch = x_batch.astype('float32')  
y_batch = y_batch.astype('float32')  
aux_batch = aux_batch.astype('float32')  
  
x_batch = torch.from_numpy(x_batch).to(device)  
aux_batch = torch.from_numpy(aux_batch).to(device)  
y_batch = torch.from_numpy(y_batch).to(device)

就可以正常跑了