哪些变量可放到GPU上-to(device)

# 参考资料：https://blog.csdn.net/qq_38101208/article/details/110481390
#%%
import torch
import torch.nn as nn

device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')

class CNN(nn.Module):
    def __init__(self):
        super(CNN,self).__init__()
        pass

#网络
model = CNN().to(device)

#训练时(输入训练数据，标签)
x = torch.arange(4)
y = torch.arange(4)

x,y = x.to(device),y.to(device)

#预测时(输入训练数据)
#输出结果如果需要用numpy 进行处理，需要讲结果载入到CPU上
out = model(x).cpu().numpy()