diff --git a/gpu_mnist_example/train_gpu.py b/gpu_mnist_example/train_gpu.py
index f163b42..7bdebcc 100644
--- a/gpu_mnist_example/train_gpu.py
+++ b/gpu_mnist_example/train_gpu.py
@@ -89,14 +89,14 @@ if __name__ == '__main__':
     device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
     batch_size = args.batch_size
     epochs = args.epoch_size
-    train_dataset = mnist.MNIST(root=os.path.join(dataset_path, "train"), train=True, transform=ToTensor(),download=False)
-    test_dataset = mnist.MNIST(root=os.path.join(dataset_path, "test"), train=False, transform=ToTensor(),download=False)
+    train_dataset = mnist.MNIST(root=os.path.join(dataset_path + "/MnistDataset_torch", "train"), train=True, transform=ToTensor(),download=False)
+    test_dataset = mnist.MNIST(root=os.path.join(dataset_path+ "/MnistDataset_torch", "test"), train=False, transform=ToTensor(),download=False)
     train_loader = DataLoader(train_dataset, batch_size=batch_size)
     test_loader = DataLoader(test_dataset, batch_size=batch_size)
 
     #如果有保存的模型，则加载模型，并在其基础上继续训练
-    if os.path.exists(os.path.join(pretrain_model_path, "mnist_epoch1_0.76.pkl")):
-        checkpoint = torch.load(os.path.join(pretrain_model_path, "mnist_epoch1_0.76.pkl"))
+    if os.path.exists(os.path.join(pretrain_model_path + "/MNIST_Example_model_zjdt", "mnist_epoch1_0.76.pkl")):
+        checkpoint = torch.load(os.path.join(pretrain_model_path + "/MNIST_Example_model_zjdt", "mnist_epoch1_0.76.pkl"))
         model.load_state_dict(checkpoint['model'])
         optimizer.load_state_dict(checkpoint['optimizer'])
         start_epoch = checkpoint['epoch']