Make sure lod of state kept consistent.

fluid/rnn_beam_search/beam_search_api.py

@@ -195,7 +195,6 @@ class StateCell(object):
                                  'Please make sure %s in input '
                                  'place holder.' % (input_name, input_name))
             self._inputs[input_name] = input_value
-
         self._state_updater(self)
 
     def update_states(self):

fluid/rnn_beam_search/simple_seq2seq.py

@@ -64,7 +64,8 @@ def decoder_state_cell(context):
     def updater(state_cell):
         current_word = state_cell.get_input('x')
         prev_h = state_cell.get_state('h')
-        h = pd.fc(input=[current_word, prev_h], size=decoder_size, act='tanh')
+        # make sure lod of h heritted from prev_h
+        h = pd.fc(input=[prev_h, current_word], size=decoder_size, act='tanh')
         state_cell.set_state('h', h)
 
     return state_cell
@@ -101,9 +102,9 @@ def decoder_decode(state_cell):
         name="init_scores", shape=[1], dtype="float32", lod_level=2)
 
     def embedding(input):
-        pd.embedding(
+        return pd.embedding(
             input=input,
-            size=[dict_dim, word_dim],
+            size=[dict_size, word_dim],
             dtype='float32',
             is_sparse=IS_SPARSE,
             param_attr=fluid.ParamAttr('vemb'))
@@ -237,5 +238,5 @@ def decode_main():
 
 
 if __name__ == '__main__':
-    train_main()
-    #decode_main()
+    #train_main()
+    decode_main()