diff --git a/TensorFlow2/built-in/cv/image_classification/MNIST_ID2481_for_TensorFlow2.X/README.md b/TensorFlow2/built-in/cv/image_classification/MNIST_ID2481_for_TensorFlow2.X/README.md
index 42959f171f2b7d11beab319080190a2bf35cf217..837522b1ac152bb22a13f18532af3115ba6f4f60 100644
--- a/TensorFlow2/built-in/cv/image_classification/MNIST_ID2481_for_TensorFlow2.X/README.md
+++ b/TensorFlow2/built-in/cv/image_classification/MNIST_ID2481_for_TensorFlow2.X/README.md
@@ -94,7 +94,7 @@ pip3 install requirements.txt
 
 ## 数据集准备<a name="section361114841316"></a>
 
-1. 用户需自行下载ml-1m训练数据集，应有如下结构
+1. 用户需自行下载MNIST训练数据集，应有如下结构
    ```
    dataset/
    ├── mnist
diff --git a/TensorFlow2/built-in/nlp/Transformer_ID0633_for_TensorFlow2.X/test/train_full_1p_4096bs_dynamic_noeval.sh b/TensorFlow2/built-in/nlp/Transformer_ID0633_for_TensorFlow2.X/test/train_full_1p_4096bs_dynamic_noeval.sh
index 632d4525fb67313007945d08934b57988b6df877..1f48774f0634a8d2497bc06ec559a8cf74baf24b 100644
--- a/TensorFlow2/built-in/nlp/Transformer_ID0633_for_TensorFlow2.X/test/train_full_1p_4096bs_dynamic_noeval.sh
+++ b/TensorFlow2/built-in/nlp/Transformer_ID0633_for_TensorFlow2.X/test/train_full_1p_4096bs_dynamic_noeval.sh
@@ -4,7 +4,8 @@
 cur_path=`pwd`
 
 #集合通信参数,不需要修改
-
+export DUMP_GE_GRAPH=2
+export DUMP_GRAPH_LEVEL=3
 export RANK_SIZE=1
 export JOB_ID=10087
 export RANK_ID_START=0
@@ -22,7 +23,7 @@ Network="Transformer_ID0633_for_TensorFlow2.X"
 #训练batch_size
 batch_size=4096
 #训练step
-train_steps=400000
+train_steps=300
 
 #TF2.X独有，不需要修改
 #export NPU_ENABLE_PERF=true
@@ -120,7 +121,7 @@ do
 	--param_set=big \
 	--train_steps=${train_steps} \
 	--batch_size=${batch_size} \
-	--steps_between_evals=10000 \
+	--steps_between_evals=100 \
 	--max_length=64 \
 	--mode=train \
 	--decode_batch_size=32 \
@@ -130,7 +131,7 @@ do
 	--dtype=fp16 \
 	--distribution_strategy='one_device' \
 	--enable_time_history=true \
-  --log_steps=1000 \
+  --log_steps=100 \
   --loss_scale='dynamic' \
 	--precision_mode=${precision_mode} \
         --over_dump=${over_dump} \
diff --git a/TensorFlow2/built-in/nlp/Word2vec_ID2350_for_TensorFlow2.X/README.md b/TensorFlow2/built-in/nlp/Word2vec_ID2350_for_TensorFlow2.X/README.md
index 3787419458b2ba009e89de75373aa590acddbdb1..be1ce02464570b74499671e9e357c7fe78207350 100644
--- a/TensorFlow2/built-in/nlp/Word2vec_ID2350_for_TensorFlow2.X/README.md
+++ b/TensorFlow2/built-in/nlp/Word2vec_ID2350_for_TensorFlow2.X/README.md
@@ -32,7 +32,8 @@
 
 ## 简述<a name="section194554031510"></a>
 
-本项目是基于TensorFlow2.X的文本分类任务，通过直接配置可以支持：TextCNN/TextRNN/TextRCNN/Transformer/Bert/Albert/DistilBert基本分类模型；TextCNN/TextRNN/TextRCNN/Transformer的token可选用词粒度/子粒度；Word2Vec特征增强后接TextCNN/TextRNN/TextRCNN/Transformer；支持Attention-TextCNN/TextRNN；FGM和PGD两种对抗方法的引入训练；对比学习方法R-drop引入；支持二分类和多分类，支持FocalLoss；保存为pb文件可供部署；项目代码支持交互式测试和批量测试。
+本项目是基于TensorFlow2.X的文本分类任务，使用Word2vec词向量训练模型进行文本分类。Word2vec是一种将词转化成向量的方法，其中包含两种算法，分别是skip-gram和CBOW，它们最大的区别是skip-gram是通过中心词去预测中心词周围的词，而CBOW是通过周围的词去预测中心词。
+
 
 - 参考论文：