diff --git a/PaddleNLP/Research/MRQA2019-D-NET/knowledge_distillation/README.md b/PaddleNLP/Research/MRQA2019-D-NET/knowledge_distillation/README.md
index 0d0b217cc9b112bfe97a4bc17946638e433caa9d..97cf9da6cc74cec25bdbda9c2c3e8991f65cea0d 100644
--- a/PaddleNLP/Research/MRQA2019-D-NET/knowledge_distillation/README.md
+++ b/PaddleNLP/Research/MRQA2019-D-NET/knowledge_distillation/README.md
@@ -30,7 +30,7 @@ models:
 
 ./data/pretrain_model/squad2_model: pretrain model(google squad2.0 model as pretrain model [Model Link](https://worksheets.codalab.org/worksheets/0x3852e60a51d2444680606556d404c657)).
 
-./saved_models/knowledge_distillation_model: baidu trained knowledge distillation model.
+./data/saved_models/knowledge_distillation_model: baidu trained knowledge distillation model.
 
 ## 3、Train and Predict
 Train and predict  knowledge distillation model
diff --git a/PaddleNLP/Research/MRQA2019-D-NET/multi_task_learning/README.md b/PaddleNLP/Research/MRQA2019-D-NET/multi_task_learning/README.md
index 865adc4a8100761d5497b08a7c86c25631c77ed2..ffb2316de1c0e4bb3d500f567f684824217d0f99 100644
--- a/PaddleNLP/Research/MRQA2019-D-NET/multi_task_learning/README.md
+++ b/PaddleNLP/Research/MRQA2019-D-NET/multi_task_learning/README.md
@@ -9,6 +9,8 @@ We use PaddlePaddle PALM(multi-task Learning Library) to train MRQA2019 MRC mult
 git clone https://github.com/PaddlePaddle/PALM.git
 ```
 
+PALM user guide: [README.md](https://github.com/PaddlePaddle/PALM/blob/master/README.md)
+
 ### Environment
 - Python >= 2.7
 - cuda >= 9.0
diff --git a/PaddleNLP/Research/MRQA2019-D-NET/server/bert_server/pdnlp/nets/bert.py b/PaddleNLP/Research/MRQA2019-D-NET/server/bert_server/pdnlp/nets/bert.py
index 67f1f51ca232bc0aa8d96791957b9d299a8f9f95..8e604a2cf957a487a434aed6184cfc4b101725ec 100644
--- a/PaddleNLP/Research/MRQA2019-D-NET/server/bert_server/pdnlp/nets/bert.py
+++ b/PaddleNLP/Research/MRQA2019-D-NET/server/bert_server/pdnlp/nets/bert.py
@@ -22,8 +22,8 @@ import json
 import numpy as np
 import paddle.fluid as fluid
 
-from palm.nets.transformer_encoder import encoder as encoder
-from palm.nets.transformer_encoder import pre_process_layer as pre_process_layer
+from pdnlp.nets.transformer_encoder import encoder as encoder
+from pdnlp.nets.transformer_encoder import pre_process_layer as pre_process_layer
 
 
 class BertModel(object):