diff --git a/PaddleNLP/neural_machine_translation/transformer/infer.py b/PaddleNLP/neural_machine_translation/transformer/infer.py
index cb40e6851db11c54faf0a393bd26f5263b16ae65..8a6804e6a496888e503449190ddc74846b1cff6e 100644
--- a/PaddleNLP/neural_machine_translation/transformer/infer.py
+++ b/PaddleNLP/neural_machine_translation/transformer/infer.py
@@ -48,14 +48,14 @@ def parse_args():
         help="The buffer size to pool data.")
     parser.add_argument(
         "--special_token",
-        type=str,
-        default=["<s>", "<e>", "<unk>"],
+        type=lambda x: x.encode(),
+        default=[b"<s>", b"<e>", b"<unk>"],
         nargs=3,
         help="The <bos>, <eos> and <unk> tokens in the dictionary.")
     parser.add_argument(
         "--token_delimiter",
-        type=lambda x: str(x.encode().decode("unicode-escape")),
-        default=" ",
+        type=lambda x: x.encode(),
+        default=b" ",
         help="The delimiter used to split tokens in source or target sentences. "
         "For EN-DE BPE data we provided, use spaces as token delimiter. ")
     parser.add_argument(
diff --git a/PaddleNLP/neural_machine_translation/transformer/reader.py b/PaddleNLP/neural_machine_translation/transformer/reader.py
index 5cceec3377ef1ad833d87672f94bb8dd119a571b..df8a45082dad1caf3561090249dd41de0a6d2b17 100644
--- a/PaddleNLP/neural_machine_translation/transformer/reader.py
+++ b/PaddleNLP/neural_machine_translation/transformer/reader.py
@@ -183,11 +183,11 @@ class DataReader(object):
                  shuffle_seed=None,
                  shuffle_batch=False,
                  use_token_batch=False,
-                 field_delimiter="\t",
-                 token_delimiter=" ",
-                 start_mark="<s>",
-                 end_mark="<e>",
-                 unk_mark="<unk>",
+                 field_delimiter=b"\t",
+                 token_delimiter=b" ",
+                 start_mark=b"<s>",
+                 end_mark=b"<e>",
+                 unk_mark=b"<unk>",
                  seed=0):
         self._src_vocab = self.load_dict(src_vocab_fpath)
         self._only_src = True
@@ -254,9 +254,9 @@ class DataReader(object):
             if tar_fname is None:
                 raise Exception("If tar file provided, please set tar_fname.")
 
-            f = tarfile.open(fpaths[0], "r")
+            f = tarfile.open(fpaths[0], "rb")
             for line in f.extractfile(tar_fname):
-                fields = line.strip("\n").split(self._field_delimiter)
+                fields = line.strip(b"\n").split(self._field_delimiter)
                 if (not self._only_src and len(fields) == 2) or (
                         self._only_src and len(fields) == 1):
                     yield fields
@@ -267,9 +267,7 @@ class DataReader(object):
 
                 with open(fpath, "rb") as f:
                     for line in f:
-                        if six.PY3:
-                            line = line.decode("utf8", errors="ignore")
-                        fields = line.strip("\n").split(self._field_delimiter)
+                        fields = line.strip(b"\n").split(self._field_delimiter)
                         if (not self._only_src and len(fields) == 2) or (
                                 self._only_src and len(fields) == 1):
                             yield fields
@@ -279,12 +277,10 @@ class DataReader(object):
         word_dict = {}
         with open(dict_path, "rb") as fdict:
             for idx, line in enumerate(fdict):
-                if six.PY3:
-                    line = line.decode("utf8", errors="ignore")
                 if reverse:
-                    word_dict[idx] = line.strip("\n")
+                    word_dict[idx] = line.strip(b"\n")
                 else:
-                    word_dict[line.strip("\n")] = idx
+                    word_dict[line.strip(b"\n")] = idx
         return word_dict
 
     def batch_generator(self):
diff --git a/PaddleNLP/neural_machine_translation/transformer/train.py b/PaddleNLP/neural_machine_translation/transformer/train.py
index fd27820bd431799b57e3c082fd14407e1d1d48e9..09c50e453fd5b5bb7b25a69e65deb83c5e91b62f 100644
--- a/PaddleNLP/neural_machine_translation/transformer/train.py
+++ b/PaddleNLP/neural_machine_translation/transformer/train.py
@@ -86,14 +86,14 @@ def parse_args():
         help="The flag indicating whether to shuffle the data batches.")
     parser.add_argument(
         "--special_token",
-        type=str,
-        default=["<s>", "<e>", "<unk>"],
+        type=lambda x: x.encode(),
+        default=[b"<s>", b"<e>", b"<unk>"],
         nargs=3,
         help="The <bos>, <eos> and <unk> tokens in the dictionary.")
     parser.add_argument(
         "--token_delimiter",
-        type=lambda x: str(x.encode().decode("unicode-escape")),
-        default=" ",
+        type=lambda x: x.encode(),
+        default=b" ",
         help="The delimiter used to split tokens in source or target sentences. "
         "For EN-DE BPE data we provided, use spaces as token delimiter. ")
     parser.add_argument(
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
index cf89607d7b103cbe65f28868a6ce6f92691f618a..9f168199b7a89697489c047d0e3b724b187421fc 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
@@ -46,14 +46,14 @@ def parse_args():
         help="The buffer size to pool data.")
     parser.add_argument(
         "--special_token",
-        type=str,
-        default=["<s>", "<e>", "<unk>"],
+        type=lambda x: x.encode(),
+        default=[b"<s>", b"<e>", b"<unk>"],
         nargs=3,
         help="The <bos>, <eos> and <unk> tokens in the dictionary.")
     parser.add_argument(
         "--token_delimiter",
-        type=lambda x: str(x.encode().decode("unicode-escape")),
-        default=" ",
+        type=lambda x: x.encode(),
+        default=b" ",
         help="The delimiter used to split tokens in source or target sentences. "
         "For EN-DE BPE data we provided, use spaces as token delimiter. ")
     parser.add_argument(
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
index 0a846825e0027d2fef9ee4515a0ac2b887806c1c..923e818717d0d90548afb35177a58f461c27ec9c 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
@@ -182,11 +182,11 @@ class DataReader(object):
                  shuffle=True,
                  shuffle_batch=False,
                  use_token_batch=False,
-                 field_delimiter="\t",
-                 token_delimiter=" ",
-                 start_mark="<s>",
-                 end_mark="<e>",
-                 unk_mark="<unk>",
+                 field_delimiter=b"\t",
+                 token_delimiter=b" ",
+                 start_mark=b"<s>",
+                 end_mark=b"<e>",
+                 unk_mark=b"<unk>",
                  seed=0):
         self._src_vocab = self.load_dict(src_vocab_fpath)
         self._only_src = True
@@ -252,9 +252,9 @@ class DataReader(object):
             if tar_fname is None:
                 raise Exception("If tar file provided, please set tar_fname.")
 
-            f = tarfile.open(fpaths[0], "r")
+            f = tarfile.open(fpaths[0], "rb")
             for line in f.extractfile(tar_fname):
-                fields = line.strip("\n").split(self._field_delimiter)
+                fields = line.strip(b"\n").split(self._field_delimiter)
                 if (not self._only_src and len(fields) == 2) or (
                         self._only_src and len(fields) == 1):
                     yield fields
@@ -265,9 +265,7 @@ class DataReader(object):
 
                 with open(fpath, "rb") as f:
                     for line in f:
-                        if six.PY3:
-                            line = line.decode("utf8", errors="ignore")
-                        fields = line.strip("\n").split(self._field_delimiter)
+                        fields = line.strip(b"\n").split(self._field_delimiter)
                         if (not self._only_src and len(fields) == 2) or (
                                 self._only_src and len(fields) == 1):
                             yield fields
@@ -277,12 +275,10 @@ class DataReader(object):
         word_dict = {}
         with open(dict_path, "rb") as fdict:
             for idx, line in enumerate(fdict):
-                if six.PY3:
-                    line = line.decode("utf8", errors="ignore")
                 if reverse:
-                    word_dict[idx] = line.strip("\n")
+                    word_dict[idx] = line.strip(b"\n")
                 else:
-                    word_dict[line.strip("\n")] = idx
+                    word_dict[line.strip(b"\n")] = idx
         return word_dict
 
     def batch_generator(self):
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/train.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/train.py
index 4313f8b441ee194935c7c47abc52271589c7765d..ade645997c8698c783372cf1cfd7c87a2c67effa 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/train.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/train.py
@@ -74,14 +74,14 @@ def parse_args():
         help="The flag indicating whether to shuffle the data batches.")
     parser.add_argument(
         "--special_token",
-        type=str,
-        default=["<s>", "<e>", "<unk>"],
+        type=lambda x: x.encode(),
+        default=[b"<s>", b"<e>", b"<unk>"],
         nargs=3,
         help="The <bos>, <eos> and <unk> tokens in the dictionary.")
     parser.add_argument(
         "--token_delimiter",
-        type=lambda x: str(x.encode().decode("unicode-escape")),
-        default=" ",
+        type=lambda x: x.encode(),
+        default=b" ",
         help="The delimiter used to split tokens in source or target sentences. "
         "For EN-DE BPE data we provided, use spaces as token delimiter. ")
     parser.add_argument(