diff --git a/README.md b/README.md
index ef89602d2919a0aaac5134519e33b110b7c743e8..e1bd0e0ca79ed0a8be6fec793961b63593d3dfa4 100644
--- a/README.md
+++ b/README.md
@@ -103,8 +103,8 @@ You can easily re-produce following competitive results with minor codes, which
       <td>94.9</td>
       <td>94.5</td>
       <td>94.7</td>
-      <td>96.3</td>
-      <td>84.0</td>
+      <td>64.3</td>
+      <td>85.2</td>
     </tr>
 
   </tbody>
diff --git a/examples/mrc/README.md b/examples/mrc/README.md
index 2f93802ea60b9da8fa5a7b12e757fc42a3d6c990..9d40fd0effbd278a7824aa25f716402eafc5c95c 100644
--- a/examples/mrc/README.md
+++ b/examples/mrc/README.md
@@ -94,5 +94,5 @@ The evaluation results are as follows:
 
 ```
 data_num: 3219
-em_sroce: 0.963031997515, f1: 83.9865402973
+em_sroce: 64.3367505436, f1: 85.1781896843
 ```
diff --git a/examples/mrc/run.py b/examples/mrc/run.py
index 838585073dbbf55ff1703c83441f8fa875fec900..fc3ee79ca1c16d2d574a49a5027eae624085a55b 100644
--- a/examples/mrc/run.py
+++ b/examples/mrc/run.py
@@ -9,7 +9,7 @@ if __name__ == '__main__':
     # configs
     max_seqlen = 512
     batch_size = 8   
-    num_epochs = 8
+    num_epochs = 2
     lr = 3e-5
     doc_stride = 128
     max_query_len = 64
@@ -64,8 +64,7 @@ if __name__ == '__main__':
     # step 8-1*: load pretrained parameters
     trainer.load_pretrain(pre_params)
     # step 8-2*: set saver to save model
-    # save_steps = (n_steps-8)  // 4
-    save_steps = 1520
+    save_steps = 3040
     trainer.set_saver(save_path=save_path, save_steps=save_steps, save_type=save_type)
     # step 8-3: start training
     trainer.train(print_steps=print_steps)
@@ -90,7 +89,7 @@ if __name__ == '__main__':
     trainer.build_predict_forward(pred_ernie, mrc_pred_head)
 
     # step 6: load checkpoint
-    pred_model_path =  './outputs/ckpt.step'+str(12160)
+    pred_model_path =  './outputs/ckpt.step'+str(3040)
     trainer.load_ckpt(pred_model_path)
     
     # step 7: fit prepared reader and data