diff --git a/tutorials/source_zh_cn/advanced_use/images/cloud_train_job1.png b/tutorials/source_zh_cn/advanced_use/images/cloud_train_job1.png
index 9ab69c08ceb9c29c90ae5d76795d9672e55f9bc2..9feb0e5388843e7d2005d9fd9c3b8dc8a2f4b17e 100644
Binary files a/tutorials/source_zh_cn/advanced_use/images/cloud_train_job1.png and b/tutorials/source_zh_cn/advanced_use/images/cloud_train_job1.png differ
diff --git a/tutorials/source_zh_cn/advanced_use/images/cloud_train_job2.png b/tutorials/source_zh_cn/advanced_use/images/cloud_train_job2.png
index e487fa0a075301505693c0bc73554e75b952cfad..9932bc8e141d95dcaa986f78afac09833d0a43c5 100644
Binary files a/tutorials/source_zh_cn/advanced_use/images/cloud_train_job2.png and b/tutorials/source_zh_cn/advanced_use/images/cloud_train_job2.png differ
diff --git a/tutorials/source_zh_cn/advanced_use/images/train_log_1_Ascend.png b/tutorials/source_zh_cn/advanced_use/images/train_log_1_Ascend.png
index 826646420b0b1b1b58907add4bcd36ed48d086ad..c44f557359ec6340b0009e7935aefe81df3ee3b2 100644
Binary files a/tutorials/source_zh_cn/advanced_use/images/train_log_1_Ascend.png and b/tutorials/source_zh_cn/advanced_use/images/train_log_1_Ascend.png differ
diff --git a/tutorials/source_zh_cn/advanced_use/images/train_log_8_Ascend.png b/tutorials/source_zh_cn/advanced_use/images/train_log_8_Ascend.png
index 10431bb60707810da54e9ed32cf2c0e1aa97c453..b2b7429c2fd169b92b18968fff4f8c07802d6a52 100644
Binary files a/tutorials/source_zh_cn/advanced_use/images/train_log_8_Ascend.png and b/tutorials/source_zh_cn/advanced_use/images/train_log_8_Ascend.png differ
diff --git a/tutorials/source_zh_cn/advanced_use/images/train_log_8_Ascend_clu.png b/tutorials/source_zh_cn/advanced_use/images/train_log_8_Ascend_clu.png
new file mode 100644
index 0000000000000000000000000000000000000000..fb24ae6e349ac3a8891ebc0f8610f46682421232
Binary files /dev/null and b/tutorials/source_zh_cn/advanced_use/images/train_log_8_Ascend_clu.png differ
diff --git a/tutorials/source_zh_cn/advanced_use/use_on_the_cloud.md b/tutorials/source_zh_cn/advanced_use/use_on_the_cloud.md
index d648ced35b10c99d3226a1768a1b571a06fdd4bc..539af81231069237605566debfa5cbd18f568188 100644
--- a/tutorials/source_zh_cn/advanced_use/use_on_the_cloud.md
+++ b/tutorials/source_zh_cn/advanced_use/use_on_the_cloud.md
@@ -269,7 +269,7 @@ if __name__ == '__main__':
 ## 创建训练任务
 
 准备好数据和执行脚本以后，需要创建训练任务将MindSpore脚本真正运行起来。首次使用ModelArts的用户可以根据本章节了解ModelArts创建训练作业的流程。
-> 本章节以MindSpore 0.1.0-alpha版本为例。
+> 本章节以MindSpore 0.5.0-alpha版本为例。ModelArts支持的MindSpore版本以创建训练作业时"算法来源"->"常用框架"->"AI引擎"可选择的版本为准。
 
 ### 进入ModelArts控制台
 
@@ -283,7 +283,7 @@ ModelArts教程<https://support.huaweicloud.com/engineers-modelarts/modelarts_23
 
 以本教程使用的训练脚本和数据为例，详细列出在创建训练作业界面如何进行配置：
 
-1. `算法来源`选择`常用框架 > Ascend-Powered-Engine > MindSpore-0.1-aarch64-cp37`。
+1. `算法来源`选择`常用框架 > Ascend-Powered-Engine > MindSpore-0.5-python3.7-aarch64`。
 
 2. `代码目录`选择预先在OBS桶中创建代码目录，`启动文件`选择代码目录下的启动脚本。
 
@@ -305,11 +305,14 @@ ModelArts教程<https://support.huaweicloud.com/engineers-modelarts/modelarts_23
 
 1. 在训练作业界面可以查看运行日志
 
-    采用`8*Ascend`规格执行ResNet-50训练任务，epoch总数为90，精度约为91%，每秒训练图片张数约12200，日志如下图所示。
+    采用`8*Ascend`规格执行ResNet-50训练任务，epoch总数为92，精度约为92%，每秒训练图片张数约12000，日志如下图所示。
+
+    ![8*Ascend训练执行结果](./images/train_log_8_Ascend_clu.png)
 
     ![8*Ascend训练执行结果](./images/train_log_8_Ascend.png)
 
-    采用`1*Ascend`规格执行ResNet-50训练任务。epoch总数为90，精度约为91%，每秒训练图片张数约1600，日志如下图所示。
+
+    采用`1*Ascend`规格执行ResNet-50训练任务。epoch总数为92，精度约为95%，每秒训练图片张数约1800，日志如下图所示。
 
     ![1*Ascend训练执行结果](./images/train_log_1_Ascend.png)
 
diff --git a/tutorials/tutorial_code/sample_for_cloud/resnet50_train.py b/tutorials/tutorial_code/sample_for_cloud/resnet50_train.py
index 57c747d559e3a56ceb2db15871599329431c9ccd..a88311466e7de4ee0e4a176d4dc46f1385c716e3 100644
--- a/tutorials/tutorial_code/sample_for_cloud/resnet50_train.py
+++ b/tutorials/tutorial_code/sample_for_cloud/resnet50_train.py
@@ -140,11 +140,15 @@ def resnet50_train(args_opt):
 
     # create model
     net = resnet50(class_num = class_num)
+    # reduction='mean' means that apply reduction of mean to loss
     loss = SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
     lr = Tensor(get_lr(global_step=0, total_epochs=epoch_size, steps_per_epoch=train_step_size))
     opt = Momentum(net.trainable_params(), lr, momentum=0.9, weight_decay=1e-4, loss_scale=loss_scale_num)
     loss_scale = FixedLossScaleManager(loss_scale_num, False)
 
+    # amp_level="O2" means that the hybrid precision of O2 mode is used for training
+    # the whole network except that batchnoram will be cast into float16 format and dynamic loss scale will be used
+    # 'keep_batchnorm_fp32 = False' means that use the float16 format
     model = Model(net, amp_level="O2", keep_batchnorm_fp32=False, loss_fn=loss, optimizer=opt, loss_scale_manager=loss_scale, metrics={'acc'})
 
     # define performance callback to show ips and loss callback to show loss for every epoch