diff --git a/paddle/fluid/operators/sequence_expand_op.cc b/paddle/fluid/operators/sequence_expand_op.cc
index d4bf6034ed56f922f70ba90dfa3ea1a9ec91d319..786fe63e7580ce16b946d5049a490eed2c3c6ced 100644
--- a/paddle/fluid/operators/sequence_expand_op.cc
+++ b/paddle/fluid/operators/sequence_expand_op.cc
@@ -145,7 +145,7 @@ and input(Y)
              [0, 3, 6, 6, 8]]
 ref_level: 0
 then we get 1-level LoDTensor
-    Out.lod =  [[0,   2,             5,             8]]
+    Out.lod =  [[0,   1,   2,        5,             8]]
     Out.data = [[a], [a], [b], [c], [d], [b], [c], [d]]
     Out.dims = [8, 1]
 
@@ -157,7 +157,7 @@ Given a common Tensor input(X)
 and input(Y)
     Y.lod = [[0, 2, 3, 6]]
 ref_level: -1
-then we a common Tensor
+then we get a common Tensor
     Out.data = [[a], [a], [b], [c], [c], [c]]
     Out.dims = [6, 1]
 
diff --git a/paddle/fluid/operators/sequence_expand_op.h b/paddle/fluid/operators/sequence_expand_op.h
index eea3cf0440fdde99d3086c06bcba9111666de0be..db7d8bd6821fabd9714a160970558291ec47197f 100644
--- a/paddle/fluid/operators/sequence_expand_op.h
+++ b/paddle/fluid/operators/sequence_expand_op.h
@@ -37,13 +37,6 @@ class SequenceExpandKernel : public framework::OpKernel<T> {
     int ref_level = context.Attr<int>("ref_level");
     auto& x_lod = x->lod();
     auto& y_lod = y->lod();
-    PADDLE_ENFORCE_GT(y_lod.size(), 0,
-                      "Level number of `Y`'s lod should be greater than 0.");
-    PADDLE_ENFORCE(
-        ref_level == -1 || (ref_level >= 0 && ref_level < y_lod.size()),
-        "Invlid `ref_level`, which should be either equal to -1 "
-        "or in [0, %d)",
-        y_lod.size());
 
     if (ref_level == -1) ref_level = y_lod.size() - 1;
 
diff --git a/python/paddle/fluid/tests/book/test_machine_translation.py b/python/paddle/fluid/tests/book/test_machine_translation.py
index fa38bd3762423497b82c3b421b3a1db4cd87525b..3a1a0859ecfd4ac5337e2112f8b22e32d8474f22 100644
--- a/python/paddle/fluid/tests/book/test_machine_translation.py
+++ b/python/paddle/fluid/tests/book/test_machine_translation.py
@@ -118,12 +118,12 @@ def decoder_decode(context, is_sparse):
             is_sparse=is_sparse)
 
         # use rnn unit to update rnn
-        current_state = pd.fc(input=[pre_ids_emb, pre_state_expanded],
+        current_state = pd.fc(input=[pre_state_expanded, pre_ids_emb],
                               size=decoder_size,
                               act='tanh')
-
+        current_state_with_lod = pd.lod_reset(x=current_state, y=pre_score)
         # use score to do beam search
-        current_score = pd.fc(input=current_state,
+        current_score = pd.fc(input=current_state_with_lod,
                               size=target_dict_dim,
                               act='softmax')
         topk_scores, topk_indices = pd.topk(current_score, k=50)