Merge pull request PaddlePaddle#77 from chengduoZH/fix_PaddingRNN

Fix bug of PaddingRnn/lm_model.py

Author: chengduo

PaddingRNN/lstm_paddle/lm_model.py

@@ -246,13 +246,13 @@ def encoder_static(input_embedding, len=3, init_hidden=None,
     init_cell = layers.data(name="init_cell", shape=[num_layers, batch_size, hidden_size], 
             dtype='float32', append_batch_size=False)
 
+    init_cell.persistable = True
+    init_hidden.persistable = True
+
     init_hidden = layers.reshape(
         init_hidden, shape=[num_layers, -1, hidden_size])
     init_cell = layers.reshape(init_cell, shape=[num_layers, -1, hidden_size])
 
-    init_cell.persistable = True
-    init_hidden.persistable = True
-
     x_emb = layers.embedding(
         input=x,
         size=[vocab_size, hidden_size],