diff --git a/PaddleCV/PaddleDetection/slim/distillation/README.md b/PaddleCV/PaddleDetection/slim/distillation/README.md
index de6a1eef06405626d2e570c9f964e6334b76dd20..88d753bbf3686ec8fd6a9c623607918c70584328 100755
--- a/PaddleCV/PaddleDetection/slim/distillation/README.md
+++ b/PaddleCV/PaddleDetection/slim/distillation/README.md
@@ -87,6 +87,16 @@ strategies:
 
 如果不需要保存评估模型，可以在定义Compressor对象时，将`save_eval_model`选项设置为False（默认为True）。
 
+运行命令为：
+```
+python ../eval.py \
+    --model_path ${checkpoint_path}/${epoch_id}/eval_model/ \
+    --model_name __model__ \
+    --params_name __params__ \
+    -c ../../configs/yolov3_mobilenet_v1_voc.yml \
+    -d "../../dataset/voc"
+```
+
 ## 预测
 
 如果在配置文件中设置了`checkpoint_path`，并且在定义Compressor对象时指定了`prune_infer_model`选项，则每个epoch都会
@@ -101,6 +111,16 @@ strategies:
 
 在脚本<a href="../infer.py">slim/infer.py</a>中展示了如何使用fluid python API加载使用预测模型进行预测。
 
+运行命令为：
+```
+python ../infer.py \
+    --model_path ${checkpoint_path}/${epoch_id}/eval_model/ \
+    --model_name __model__ \
+    --params_name __params__ \
+    -c ../../configs/yolov3_mobilenet_v1_voc.yml \
+    --infer_dir ../../demo
+```
+
 ### PaddleLite
 
 该示例中产出的预测（inference）模型可以直接用PaddleLite进行加载使用。
diff --git a/PaddleCV/PaddleDetection/slim/prune/README.md b/PaddleCV/PaddleDetection/slim/prune/README.md
index 22ebbd625c63094089c7bcbc483aa3c006dbe724..ee98765843e79086c17f8066497291d2f7ef426e 100644
--- a/PaddleCV/PaddleDetection/slim/prune/README.md
+++ b/PaddleCV/PaddleDetection/slim/prune/README.md
@@ -30,7 +30,7 @@ from paddle.fluid.framework import IrGraph
 from paddle.fluid import core
 
 graph = IrGraph(core.Graph(train_prog.desc), for_test=True)
-marked_nodes = set() 
+marked_nodes = set()
 for op in graph.all_op_nodes():
     print(op.name())
     if op.name().find('conv') > -1:
@@ -40,12 +40,12 @@ graph.draw('.', 'forward', marked_nodes)
 
 该示例中MobileNetV1-YoloV3模型结构的可视化结果：<a href="./images/MobileNetV1-YoloV3.pdf">MobileNetV1-YoloV3.pdf</a>
 
-同时通过以下命令观察目标卷积层的参数（parameters）的名称和shape: 
+同时通过以下命令观察目标卷积层的参数（parameters）的名称和shape:
 
 ```
 for param in fluid.default_main_program().global_block().all_parameters():
     if 'weights' in param.name:
-        print param.name, param.shape
+        print(param.name, param.shape)
 ```
 
 
@@ -121,7 +121,7 @@ python compress.py \
 - **max_iters:** 一个`epoch`中batch的数量，需要设置为`total_num / batch_size`, 其中`total_num`为训练样本总数量，`batch_size`为多卡上总的batch size.
 - **YoloTrainFeed.batch_size:** 当使用DataLoader时，表示单张卡上的batch size; 当使用普通reader时，则表示多卡上的总的`batch_size`。`batch_size`受限于显存大小。
 - **LeaningRate.base_lr:** 根据多卡的总`batch_size`调整`base_lr`，两者大小正相关，可以简单的按比例进行调整。
-- **LearningRate.schedulers.PiecewiseDecay.milestones：**请根据batch size的变化对其调整。
+- **LearningRate.schedulers.PiecewiseDecay.milestones：** 请根据batch size的变化对其调整。
 - **LearningRate.schedulers.PiecewiseDecay.LinearWarmup.steps：** 请根据batch size的变化对其进行调整。
 
 
@@ -168,6 +168,16 @@ python compress.py \
 
 如果不需要保存评估模型，可以在定义Compressor对象时，将`save_eval_model`选项设置为False（默认为True）。
 
+运行命令为：
+```
+python ../eval.py \
+    --model_path ${checkpoint_path}/${epoch_id}/eval_model/ \
+    --model_name __model__ \
+    --params_name __params__ \
+    -c ../../configs/yolov3_mobilenet_v1_voc.yml \
+    -d "../../dataset/voc"
+```
+
 ## 预测
 
 如果在配置文件中设置了`checkpoint_path`，并且在定义Compressor对象时指定了`prune_infer_model`选项，则每个epoch都会
@@ -182,6 +192,16 @@ python compress.py \
 
 在脚本<a href="../infer.py">PaddleDetection/tools/infer.py</a>中展示了如何使用fluid python API加载使用预测模型进行预测。
 
+运行命令为：
+```
+python ../infer.py \
+    --model_path ${checkpoint_path}/${epoch_id}/eval_model/ \
+    --model_name __model__ \
+    --params_name __params__ \
+    -c ../../configs/yolov3_mobilenet_v1_voc.yml \
+    --infer_dir ../../demo
+```
+
 ### PaddleLite
 
 该示例中产出的预测（inference）模型可以直接用PaddleLite进行加载使用。
diff --git a/PaddleSlim/classification/distillation/README.md b/PaddleSlim/classification/distillation/README.md
index 1d5004699a060320ed5b08287e4e644476397212..6a097851cdb65ef27e7e0fc74419533ecc816606 100755
--- a/PaddleSlim/classification/distillation/README.md
+++ b/PaddleSlim/classification/distillation/README.md
@@ -34,12 +34,12 @@
 ```python
 # 观察student model的Variable
 for v in fluid.default_main_program().list_vars():
-    print v.name, v.shape
+    print(v.name, v.shape)
 ```
 ```python
 # 观察teacher model的Variable
 for v in teacher_program.list_vars():
-    print v.name, v.shape
+    print(v.name, v.shape)
 ```
 
 经过对比可以发现，`student model`和`teacher model`预测的输出分别为：
@@ -111,6 +111,16 @@ strategies:
 
 脚本<a href="../eval.py">PaddleSlim/classification/eval.py</a>中为使用该模型在评估数据集上做评估的示例。
 
+运行命令示例:
+```bash
+python eval.py \
+    --use_gpu True \
+    --model_path ${save_path}/eval_model/ \
+    --model_name __model__ \
+    --params_name __params__
+
+```
+
 ## 预测
 
 如果在配置文件中设置了`checkpoint_path`，并且在定义Compressor对象时指定了`prune_infer_model`选项，则每个epoch都会
@@ -125,6 +135,15 @@ strategies:
 
 在脚本<a href="../infer.py">PaddleSlim/classification/infer.py</a>中展示了如何使用fluid python API加载使用预测模型进行预测。
 
+运行命令示例:
+```bash
+python infer.py \
+    --use_gpu True \
+    --model_path ${save_path}/eval_model/ \
+    --model_name __model__ \
+    --params_name __params__
+```
+
 ### PaddleLite
 
 该示例中产出的预测（inference）模型可以直接用PaddleLite进行加载使用。
@@ -144,7 +163,7 @@ strategies:
 - batch size: 256
 - lr_strategy: piecewise_decay
 - step_epochs: 30, 60, 90
-- num_epochs: 120
+- num_epochs: 130
 - l2_decay: 4e-5
 - init lr: 0.1
 
@@ -160,7 +179,7 @@ strategies:
 - batch size: 256
 - lr_strategy: piecewise_decay
 - step_epochs: 30, 60, 90
-- num_epochs: 120
+- num_epochs: 130
 - l2_decay: 4e-5
 - init lr: 0.1
 
@@ -176,7 +195,7 @@ strategies:
 - batch size: 256
 - lr_strategy: piecewise_decay
 - step_epochs: 30, 60, 90
-- num_epochs: 120
+- num_epochs: 130
 - l2_decay: 4e-5
 - init lr: 0.1
 
diff --git a/PaddleSlim/classification/distillation/compress.py b/PaddleSlim/classification/distillation/compress.py
index 49a800f8cba00abf3faad5976c8a406ed6b1fd20..3cb8f2bcb6d1df87f8631bd69f1be3b16ea6defd 100644
--- a/PaddleSlim/classification/distillation/compress.py
+++ b/PaddleSlim/classification/distillation/compress.py
@@ -95,7 +95,6 @@ def compress(args):
     train_fetch_list = [('loss', avg_cost.name)]
 
     teacher_programs = []
-    distiller_optimizer = None
 
     teacher_model = models.__dict__[args.teacher_model](prefix_name='res50')
     # define teacher program
diff --git a/PaddleSlim/classification/distillation/configs/mobilenetv2_resnet50_distillation.yaml b/PaddleSlim/classification/distillation/configs/mobilenetv2_resnet50_distillation.yaml
index 079d9231b427a8a270cfc719db68e4f17d712507..750cbae23928352fe0903f84327be0efab9b55ef 100644
--- a/PaddleSlim/classification/distillation/configs/mobilenetv2_resnet50_distillation.yaml
+++ b/PaddleSlim/classification/distillation/configs/mobilenetv2_resnet50_distillation.yaml
@@ -2,8 +2,8 @@ version: 1.0
 distillers:
     l2_distiller:
         class: 'L2Distiller'
-        teacher_feature_map: 'res50_fc_0.tmp_1'
-        student_feature_map: 'fc_0.tmp_1'
+        teacher_feature_map: 'res50_fc_0.tmp_0'
+        student_feature_map: 'fc_0.tmp_0'
         distillation_loss_weight: 1
 strategies:
     distillation_strategy:
diff --git a/PaddleSlim/classification/distillation/run.sh b/PaddleSlim/classification/distillation/run.sh
index c18e6eeed028aa4f95735decf9901afde2e8af1a..8668c5fb4df455308c0490a88afeb32f209180e6 100644
--- a/PaddleSlim/classification/distillation/run.sh
+++ b/PaddleSlim/classification/distillation/run.sh
@@ -58,7 +58,7 @@ tailf mobilenet_v1.log
 #--model "MobileNetV2" \
 #--teacher_model "ResNet50" \
 #--teacher_pretrained_model ../pretrain/ResNet50_pretrained \
-#--compress_config ./configs/mobilenetv2_resnet50_distillation.yaml\
+#--compress_config ./configs/mobilenetv2_resnet50_distillation.yaml \
 #> mobilenet_v2.log 2>&1 &
 #tailf mobilenet_v2.log
 
diff --git a/PaddleSlim/classification/pruning/README.md b/PaddleSlim/classification/pruning/README.md
index 92082098b4e19e99c8ca9be4aa489515dbfd8394..6c4035d5d94e17e162ef95427bcbe49605c17620 100644
--- a/PaddleSlim/classification/pruning/README.md
+++ b/PaddleSlim/classification/pruning/README.md
@@ -37,7 +37,7 @@ PaddleSlim暂时无法对`depthwise convolution`直接进行剪裁， 因为`dep
 ```
 for param in fluid.default_main_program().global_block().all_parameters():
     if 'weights' in param.name:
-        print param.name, param.shape
+        print(param.name, param.shape)
 ```
 
 结果如下：
@@ -100,6 +100,16 @@ fc10_weights (1280L, 1000L)
 
 脚本<a href="../eval.py">PaddleSlim/classification/eval.py</a>中为使用该模型在评估数据集上做评估的示例。
 
+运行命令示例:
+```bash
+python eval.py \
+    --use_gpu True \
+    --model_path ${save_path}/eval_model/ \
+    --model_name __model__ \
+    --params_name __params__
+
+```
+
 ## 预测
 
 如果在配置文件中设置了`checkpoint_path`，并且在定义Compressor对象时指定了`prune_infer_model`选项，则每个epoch都会
@@ -114,6 +124,15 @@ fc10_weights (1280L, 1000L)
 
 在脚本<a href="../infer.py">PaddleSlim/classification/infer.py</a>中展示了如何使用fluid python API加载使用预测模型进行预测。
 
+运行命令示例:
+```bash
+python infer.py \
+    --use_gpu True \
+    --model_path ${save_path}/eval_model/ \
+    --model_name __model__ \
+    --params_name __params__
+```
+
 ### PaddleLite
 
 该示例中产出的预测（inference）模型可以直接用PaddleLite进行加载使用。