diff --git a/PaddleNLP/neural_machine_translation/transformer/infer.py b/PaddleNLP/neural_machine_translation/transformer/infer.py
index 08543f010163f6149c3b227ff86d0a7c95a5064c..aaf813a5e1345a47e4e9073d2ff1219274fc97fe 100644
--- a/PaddleNLP/neural_machine_translation/transformer/infer.py
+++ b/PaddleNLP/neural_machine_translation/transformer/infer.py
@@ -281,11 +281,11 @@ def fast_infer(args):
                     feed=feed_dict_list[0]
                     if feed_dict_list is not None else None,
                     return_numpy=False,
-                    use_program_cache=True)
+                    use_program_cache=False)
             seq_ids_list, seq_scores_list = [seq_ids], [
                 seq_scores
-            ] if isinstance(
-                seq_ids, paddle.fluid.LoDTensor) else (seq_ids, seq_scores)
+            ] if isinstance(seq_ids,
+                            paddle.fluid.LoDTensor) else (seq_ids, seq_scores)
             for seq_ids, seq_scores in zip(seq_ids_list, seq_scores_list):
                 # How to parse the results:
                 #   Suppose the lod of seq_ids is:
diff --git a/PaddleNLP/neural_machine_translation/transformer/reader.py b/PaddleNLP/neural_machine_translation/transformer/reader.py
index 10f44ade6768edf9536cad27bcbcd0b08d16e668..aa4dea3e380aad409d2ca30cce6658288d2c7eef 100644
--- a/PaddleNLP/neural_machine_translation/transformer/reader.py
+++ b/PaddleNLP/neural_machine_translation/transformer/reader.py
@@ -266,7 +266,7 @@ class DataReader(object):
                 with open(fpath, "rb") as f:
                     for line in f:
                         if six.PY3:
-                            line = line.decode()
+                            line = line.decode("utf8", errors="ignore")
                         fields = line.strip("\n").split(self._field_delimiter)
                         if (not self._only_src and len(fields) == 2) or (
                                 self._only_src and len(fields) == 1):
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
index 96b8e0a14fcc225e11e63a7604c8bf8e6db8e45e..cf89607d7b103cbe65f28868a6ce6f92691f618a 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
@@ -280,7 +280,7 @@ def fast_infer(args):
                     feed=feed_dict_list[0]
                     if feed_dict_list is not None else None,
                     return_numpy=False,
-                    use_program_cache=True)
+                    use_program_cache=False)
             seq_ids_list, seq_scores_list = [
                 seq_ids
             ], [seq_scores] if isinstance(
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/model.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/model.py
index cfd85dc505216a0d226a8cb29012ae3e36cc26cc..5b19be6a526852ac348d710800a741d6cfd971c1 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/model.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/model.py
@@ -875,7 +875,7 @@ def fast_decode(src_vocab_size,
             accu_scores = layers.elementwise_add(
                 x=layers.log(topk_scores), y=pre_scores, axis=0)
             # beam_search op uses lod to differentiate branches.
-            topk_indices = layers.lod_reset(accu_scores, pre_ids)
+            accu_scores = layers.lod_reset(accu_scores, pre_ids)
             # topK reduction across beams, also contain special handle of
             # end beams and end sentences(batch reduction)
             selected_ids, selected_scores, gather_idx = layers.beam_search(
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
index 10f44ade6768edf9536cad27bcbcd0b08d16e668..aa4dea3e380aad409d2ca30cce6658288d2c7eef 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
@@ -266,7 +266,7 @@ class DataReader(object):
                 with open(fpath, "rb") as f:
                     for line in f:
                         if six.PY3:
-                            line = line.decode()
+                            line = line.decode("utf8", errors="ignore")
                         fields = line.strip("\n").split(self._field_delimiter)
                         if (not self._only_src and len(fields) == 2) or (
                                 self._only_src and len(fields) == 1):