diff --git a/tutorials/source_zh_cn/advanced_use/bert_poetry.md b/tutorials/source_zh_cn/advanced_use/bert_poetry.md
index e4fc7f37ca8902a1b3d9a24647a95dc4c9cd08f8..6cdb95fe9c99617bf9603cb2a3c5309bb77b751e 100644
--- a/tutorials/source_zh_cn/advanced_use/bert_poetry.md
+++ b/tutorials/source_zh_cn/advanced_use/bert_poetry.md
@@ -81,7 +81,7 @@ BERT采用了Encoder结构，`attention_mask`为全1的向量，即每个token
 
 ![Teaser image](images/finetune.PNG)
 
-图2：训练流程示意图
+图3：训练流程示意图
 
 ## 样例代码