diff --git a/tutorials/article-implementation/ArticleReproduction_CV.md b/tutorials/article-implementation/ArticleReproduction_CV.md
index e60b1902a3837360e251e870d036301a7ce7ef0e..e9d5c76c155252def4cb1c684ea65d8897034c5a 100644
--- a/tutorials/article-implementation/ArticleReproduction_CV.md
+++ b/tutorials/article-implementation/ArticleReproduction_CV.md
@@ -277,7 +277,7 @@ if __name__ == "__main__":
 
 * 模型在前向对齐验证时，需要调用`model.eval()`方法，保证组网中的随机量被关闭，比如BatchNorm、Dropout等。
 * 给定相同的输入数据，为保证可复现性，如果有随机数生成，固定相关的随机种子。
-* 我们可以基于reprod logger 的比较结果判断对齐效果，一般误差在1e-6附近的话，可以认为前向没有问题。
+* 我们可以基于reprod logger 的比较结果判断对齐效果，一般误差在1e-5附近的话，可以认为前向没有问题。
 * 如果最终输出结果diff较大，可以使用二分的方法进行排查，比如说ResNet50，包含1个stem、4个res-stage、global avg-pooling以及最后的fc层，那么完成模型组网和权重转换之后，如果模型输出没有对齐，可以尝试输出中间某一个res-stage的tensor进行对比，如果相同，则向后进行排查；如果不同，则继续向前进行排查，以此类推，直到找到导致没有对齐的操作。
 
 **【实战】**
diff --git a/tutorials/mobilenetv3_prod/Step1-5/01_test_forward.py b/tutorials/mobilenetv3_prod/Step1-5/01_test_forward.py
index 3ce67bc727ec8dd40d97313a2e1032476ed255bf..ee4aef592660b8b14bc94c191fa0757b3c2204f5 100644
--- a/tutorials/mobilenetv3_prod/Step1-5/01_test_forward.py
+++ b/tutorials/mobilenetv3_prod/Step1-5/01_test_forward.py
@@ -9,6 +9,10 @@ from mobilenetv3_ref.torchvision.models import mobilenet_v3_small as mv3_small_t
 
 
 def test_forward():
+    device = "gpu"  # you can also set it as "cpu"
+    torch_device = torch.device("cuda:0" if device == "gpu" else "cpu")
+    paddle.set_device(device)
+
     # load paddle model
     paddle_model = mv3_small_paddle()
     paddle_model.eval()
@@ -21,6 +25,8 @@ def test_forward():
     torch_state_dict = torch.load("./data/mobilenet_v3_small-047dcff4.pth")
     torch_model.load_state_dict(torch_state_dict)
 
+    torch_model.to(torch_device)
+
     # load data
     inputs = np.load("./data/fake_data.npy")
 
@@ -31,7 +37,9 @@ def test_forward():
     reprod_logger.save("./result/forward_paddle.npy")
 
     # save the torch output
-    torch_out = torch_model(torch.tensor(inputs, dtype=torch.float32))
+    torch_out = torch_model(
+        torch.tensor(
+            inputs, dtype=torch.float32).to(torch_device))
     reprod_logger.add("logits", torch_out.cpu().detach().numpy())
     reprod_logger.save("./result/forward_ref.npy")
 
@@ -46,4 +54,5 @@ if __name__ == "__main__":
 
     # compare result and produce log
     diff_helper.compare_info(torch_info, paddle_info)
-    diff_helper.report(path="./result/log/forward_diff.log")
+    diff_helper.report(
+        path="./result/log/forward_diff.log", diff_threshold=1e-5)
diff --git a/tutorials/mobilenetv3_prod/Step1-5/README.md b/tutorials/mobilenetv3_prod/Step1-5/README.md
index ffc7bac68d46cc69e9048f41df59bf66126f3ec8..c6042ca5368f9e7c040388b49430c83101caf28e 100644
--- a/tutorials/mobilenetv3_prod/Step1-5/README.md
+++ b/tutorials/mobilenetv3_prod/Step1-5/README.md
@@ -157,14 +157,12 @@ python 01_test_forward.py
 根据示例代码可以看到，我们将结果保存在`result/log/forward_diff.log`中，打开对应文件或者直接观察命令行输出，就会有下列结果：
 
 ```bash
-[2021/12/21 15:00:38] root INFO: logits:
-[2021/12/21 15:00:38] root INFO:     mean diff: check passed: False, value: 2.308018565599923e-06
-[2021/12/21 15:00:38] root INFO: diff check failed
+[2022/02/28 05:31:40] root INFO: logits:
+[2022/02/28 05:31:40] root INFO: 	mean diff: check passed: True, value: 1.7629824924370041e-06
+[2022/02/28 05:31:40] root INFO: diff check passed
 ```
 
-这里我们发现在`reprod_log`默认的平均差异小于1e-6的标准下，当前前向对齐是不符合条件的，但是这是由于前向 op 计算导致的微小的差异。
-
-一般说来前向误差在 1e-5 左右都是可以接受的，到这里我们就验证了网络的前向是对齐的，完成了第一个打卡点。
+由于前向 op 计算导致的微小的差异。一般说来前向误差在 1e-5 左右的 diff 是可以接受的，到这里我们就验证了网络的前向是对齐的，完成了第一个打卡点。
 
 <a name="4.2"></a>
 ### 4.2 数据加载对齐
diff --git a/tutorials/mobilenetv3_prod/Step1-5/mobilenetv3_ref/presets.py b/tutorials/mobilenetv3_prod/Step1-5/mobilenetv3_ref/presets.py
index e190466cf2431cfc88b7b25506f1f212a1c7ce0b..44fcf5cc77cae62da8ffabef9c63e33c1bcbfdba 100644
--- a/tutorials/mobilenetv3_prod/Step1-5/mobilenetv3_ref/presets.py
+++ b/tutorials/mobilenetv3_prod/Step1-5/mobilenetv3_ref/presets.py
@@ -1,4 +1,4 @@
-from torchvision.transforms import autoaugment, transforms
+from torchvision.transforms import transforms
 
 
 class ClassificationPresetTrain:
diff --git a/tutorials/mobilenetv3_prod/Step1-5/result/log/forward_diff.log b/tutorials/mobilenetv3_prod/Step1-5/result/log/forward_diff.log
index e69e5d3711c73f6959b6624122e2a1ee035635fa..cd4ddae0fbee3f19c66d5b879b62c1622db12aa9 100644
--- a/tutorials/mobilenetv3_prod/Step1-5/result/log/forward_diff.log
+++ b/tutorials/mobilenetv3_prod/Step1-5/result/log/forward_diff.log
@@ -1,3 +1,3 @@
-[2022/01/03 16:50:19] root INFO: logits: 
-[2022/01/03 16:50:19] root INFO: 	mean diff: check passed: False, value: 2.308018565599923e-06
-[2022/01/03 16:50:19] root INFO: diff check failed
+[2022/02/28 05:31:40] root INFO: logits: 
+[2022/02/28 05:31:40] root INFO: 	mean diff: check passed: True, value: 1.7629824924370041e-06
+[2022/02/28 05:31:40] root INFO: diff check passed