1. torch.save

2. torch.load

方法1：保存整个module （耗时，占内存）

torch.save(net.path）
path_model = './model.pkl'
net_load = torch.load(path_model)
方法2：保存模型参数（官方推荐）
state_dict = net.state_dict()
torch.save(state_dict, path)
path_state_dict = './model_state_dict.pkl'
state_dict_load = torch.load(path_state_dict)
net.load_state_dict(state_dict_load)
3. 断点续存训练
保存断点（在epoch循环中）：
if (epoch + 1) % checkpoint_interval == 0:  # 每隔checkpoint_interval保存一次
    checkpoint = {"model_state_dict": net.state_dict()  # 模型数据
                  "optimizer_state_dict": optimizer.state_dict()  # 优化器数据
                  "epoch": epoch  # 迭代次数
    path_checkpoint = './checkpoint_{}_epoch.pkl'.format(epoch)
    torch.save(checkpoint, path_checkpoint)
断点恢复：
path_checkpoint = './checkpoint_4_epoch.pkl'
checkpoint = torch.load(path_checkpoint)