diff --git a/propeller/README.en.md b/propeller/README.en.md
index 8ced3b67fed3e997de8371d807881a888271c294..a8b9767f8ff900647d61b951776fd80fae9f0cce 100644
--- a/propeller/README.en.md
+++ b/propeller/README.en.md
@@ -21,7 +21,7 @@ Propeller provide the following benefits:
 ## install
 
 ```script
-cd propeller && pip install .
+pip install --user .
 ```
 
 ## Getting Started
@@ -71,7 +71,6 @@ cd propeller && pip install .
     # Start training!
     propeller.train_and_eval(BowModel, hparams, run_config, train_ds, eval_ds)
 ```
-More detail see example/toy/
 
 ## Main Feature
 1. train_and_eval
@@ -91,9 +90,9 @@ More detail see example/toy/
 
 4. Summary
     To trace tensor histogram in training, simply：
-    ```python
-        propeller.summary.histogram('loss', tensor) 
-    ```
+```python
+    propeller.summary.histogram('loss', tensor) 
+```
 
 
 ## Contributing
diff --git a/propeller/README.md b/propeller/README.md
index 8e1bb388616558bb7f88ffccde15b68e9c372e5f..004edadc172e578ab7f3a0381c90c207d1313545 100644
--- a/propeller/README.md
+++ b/propeller/README.md
@@ -21,7 +21,7 @@ Propeller 具有下列优势：
 
 ## install|安装
 
-cd propeller && pip install .
+pip install --user .
 
 ## Getting Started|快速开始
 ```python
@@ -70,7 +70,6 @@ cd propeller && pip install .
     # 开始训练！
     propeller.train_and_eval(BowModel, hparams, run_config, train_ds, eval_ds)
 ```
-先洗详细请见example/toy/
 
 ## 主要构件
 1. train_and_eval
@@ -89,10 +88,10 @@ cd propeller && pip install .
 
 4. Summary
     对训练过程中的某些参数进行log追踪，只需要：
-    ```python
-                propeller.summary.histogram('loss', tensor) 
+```python
+            propeller.summary.histogram('loss', tensor) 
 
-    ```
+```
 
 
 ## Contributing|贡献
diff --git a/propeller/paddle/train/metrics.py b/propeller/paddle/train/metrics.py
index 3d14659662b5ee878b84a6efe5565d0631338a83..df5ce8ef96d1ced7322f9b5af2b4ff5e1ade7af5 100644
--- a/propeller/paddle/train/metrics.py
+++ b/propeller/paddle/train/metrics.py
@@ -256,8 +256,10 @@ class Mrr(Metrics):
                     sorted(
                         tup, key=lambda t: t[2], reverse=True)) if l != 0
             ]
-            ranks = ranks[0]
-            return ranks
+            if len(ranks):
+                return ranks[0]
+            else:
+                return 0.
 
         mrr_for_qid = [
             calc_func(tup)