diff --git a/PaddleNLP/neural_machine_translation/transformer/README.md b/PaddleNLP/neural_machine_translation/transformer/README.md
index 469bea74071ec2e5712efcde0753b39b937d558d..c3d452335420ed287f1bb5ad5ae98e69723b58ef 100644
--- a/PaddleNLP/neural_machine_translation/transformer/README.md
+++ b/PaddleNLP/neural_machine_translation/transformer/README.md
@@ -55,7 +55,7 @@
 
 3. 模型预测
 
-	使用以上提供的数据和模型，可以按照以下代码进行预测，翻译结果将打印到标准输出:
+	使用以上提供的数据和模型，可以按照以下代码进行预测，翻译结果将打印到 `output_file` 指定的文件中:
 	```sh
     # base model
     python -u infer.py \
@@ -63,6 +63,7 @@
     --trg_vocab_fpath gen_data/wmt16_ende_data_bpe/vocab_all.bpe.32000 \
     --special_token '<s>' '<e>' '<unk>' \
     --test_file_pattern gen_data/wmt16_ende_data_bpe/newstest2014.tok.bpe.32000.en-de \
+    --output_file predict.txt \
     --token_delimiter ' ' \
     --batch_size 32 \
     model_path trained_models/iter_100000.infer.model \
@@ -76,6 +77,7 @@
     --trg_vocab_fpath gen_data/wmt16_ende_data_bpe/vocab_all.bpe.32000 \
     --special_token '<s>' '<e>' '<unk>' \
     --test_file_pattern gen_data/wmt16_ende_data_bpe/newstest2014.tok.bpe.32000.en-de \
+    --output_file predict.txt \
     --token_delimiter ' ' \
     --batch_size 32 \
     model_path trained_models/iter_100000.infer.model \
diff --git a/PaddleNLP/neural_machine_translation/transformer/infer.py b/PaddleNLP/neural_machine_translation/transformer/infer.py
index 9a3bfa2a78e475bd3ea996ae492bc7a08279d22a..0d53fa8f7e16ad178fb25213a48a6c148472f8dd 100644
--- a/PaddleNLP/neural_machine_translation/transformer/infer.py
+++ b/PaddleNLP/neural_machine_translation/transformer/infer.py
@@ -4,9 +4,6 @@ import multiprocessing
 import numpy as np
 import os
 import sys
-if sys.version[0] == '2':
-    reload(sys)
-    sys.setdefaultencoding("utf-8")
 sys.path.append("../../")
 sys.path.append("../../models/neural_machine_translation/transformer/")
 from functools import partial
@@ -23,7 +20,7 @@ from train import pad_batch_data, prepare_data_generator
 
 
 def parse_args():
-    parser = argparse.ArgumentParser("Training for Transformer.")
+    parser = argparse.ArgumentParser("Inference for Transformer.")
     parser.add_argument(
         "--src_vocab_fpath",
         type=str,
@@ -39,6 +36,11 @@ def parse_args():
         type=str,
         required=True,
         help="The pattern to match test data files.")
+    parser.add_argument(
+        "--output_file",
+        type=str,
+        default="predict.txt",
+        help="The file to output the translation results of to.")
     parser.add_argument(
         "--batch_size",
         type=int,
@@ -51,14 +53,14 @@ def parse_args():
         help="The buffer size to pool data.")
     parser.add_argument(
         "--special_token",
-        type=lambda x: x.encode(),
-        default=[b"<s>", b"<e>", b"<unk>"],
+        type=lambda x: x.encode("utf8"),
+        default=["<s>", "<e>", "<unk>"],
         nargs=3,
         help="The <bos>, <eos> and <unk> tokens in the dictionary.")
     parser.add_argument(
         "--token_delimiter",
-        type=lambda x: x.encode(),
-        default=b" ",
+        type=lambda x: x.encode("utf8"),
+        default=" ",
         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(
@@ -271,6 +273,7 @@ def fast_infer(args):
     trg_idx2word = reader.DataReader.load_dict(
         dict_path=args.trg_vocab_fpath, reverse=True)
 
+    f = open(args.output_file, "wb")
     while True:
         try:
             feed_dict_list = prepare_feed_dict_list(data_generator, dev_count,
@@ -316,7 +319,7 @@ def fast_infer(args):
                                 np.array(seq_ids)[sub_start:sub_end])
                         ]))
                         scores[i].append(np.array(seq_scores)[sub_end - 1])
-                        print(hyps[i][-1].decode("utf8"))
+                        f.write(hyps[i][-1] + b"\n")
                         if len(hyps[i]) >= InferTaskConfig.n_best:
                             break
         except (StopIteration, fluid.core.EOFException):
@@ -324,6 +327,7 @@ def fast_infer(args):
             if args.use_py_reader:
                 pyreader.reset()
             break
+    f.close()
 
 
 if __name__ == "__main__":
diff --git a/PaddleNLP/neural_machine_translation/transformer/reader.py b/PaddleNLP/neural_machine_translation/transformer/reader.py
index df8a45082dad1caf3561090249dd41de0a6d2b17..c1dcce6ac423565cbc634d0a7e832f60d3d0546b 100644
--- a/PaddleNLP/neural_machine_translation/transformer/reader.py
+++ b/PaddleNLP/neural_machine_translation/transformer/reader.py
@@ -183,12 +183,23 @@ class DataReader(object):
                  shuffle_seed=None,
                  shuffle_batch=False,
                  use_token_batch=False,
-                 field_delimiter=b"\t",
-                 token_delimiter=b" ",
-                 start_mark=b"<s>",
-                 end_mark=b"<e>",
-                 unk_mark=b"<unk>",
+                 field_delimiter="\t",
+                 token_delimiter=" ",
+                 start_mark="<s>",
+                 end_mark="<e>",
+                 unk_mark="<unk>",
                  seed=0):
+        # convert str to bytes, and use byte data
+        field_delimiter = field_delimiter if isinstance(
+            field_delimiter, bytes) else field_delimiter.encode("utf8")
+        token_delimiter = token_delimiter if isinstance(
+            token_delimiter, bytes) else token_delimiter.encode("utf8")
+        start_mark = start_mark if isinstance(
+            start_mark, bytes) else start_mark.encode("utf8")
+        end_mark = end_mark if isinstance(end_mark,
+                                          bytes) else end_mark.encode("utf8")
+        unk_mark = unk_mark if isinstance(unk_mark,
+                                          bytes) else unk_mark.encode("utf8")
         self._src_vocab = self.load_dict(src_vocab_fpath)
         self._only_src = True
         if trg_vocab_fpath is not None:
diff --git a/PaddleNLP/neural_machine_translation/transformer/train.py b/PaddleNLP/neural_machine_translation/transformer/train.py
index 0d30178ae98a0fc3cba1a71e7fba596aeb53201f..277cd13186065660ab484d51b91819274f1aab35 100644
--- a/PaddleNLP/neural_machine_translation/transformer/train.py
+++ b/PaddleNLP/neural_machine_translation/transformer/train.py
@@ -11,9 +11,6 @@ if os.environ.get('FLAGS_eager_delete_tensor_gb', None) is None:
 
 import six
 import sys
-if sys.version[0] == '2':
-    reload(sys)
-    sys.setdefaultencoding("utf-8")
 sys.path.append("../../")
 sys.path.append("../../models/neural_machine_translation/transformer/")
 import time
@@ -89,14 +86,14 @@ def parse_args():
         help="The flag indicating whether to shuffle the data batches.")
     parser.add_argument(
         "--special_token",
-        type=lambda x: x.encode(),
-        default=[b"<s>", b"<e>", b"<unk>"],
+        type=lambda x: x.encode("utf8"),
+        default=["<s>", "<e>", "<unk>"],
         nargs=3,
         help="The <bos>, <eos> and <unk> tokens in the dictionary.")
     parser.add_argument(
         "--token_delimiter",
-        type=lambda x: x.encode(),
-        default=b" ",
+        type=lambda x: x.encode("utf8"),
+        default=" ",
         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/README_cn.md b/PaddleNLP/unarchived/neural_machine_translation/transformer/README_cn.md
index bdac7cb0b7c4f9d51bbc281b351232c6edc75a36..a3a3bd6f9094c0912c741726ed876c72f9adb766 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/README_cn.md
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/README_cn.md
@@ -139,6 +139,7 @@ python -u infer.py \
   --trg_vocab_fpath gen_data/wmt16_ende_data_bpe/vocab_all.bpe.32000 \
   --special_token '<s>' '<e>' '<unk>' \
   --test_file_pattern gen_data/wmt16_ende_data_bpe/newstest2016.tok.bpe.32000.en-de \
+  --output_file predict.txt \
   --token_delimiter ' ' \
   --batch_size 32 \
   model_path trained_models/iter_100000.infer.model \
@@ -152,6 +153,7 @@ python -u infer.py \
   --trg_vocab_fpath gen_data/wmt16_ende_data_bpe/vocab_all.bpe.32000 \
   --special_token '<s>' '<e>' '<unk>' \
   --test_file_pattern gen_data/wmt16_ende_data_bpe/newstest2016.tok.bpe.32000.en-de \
+  --output_file predict.txt \
   --token_delimiter ' ' \
   --batch_size 32 \
   model_path trained_models/iter_100000.infer.model \
@@ -164,7 +166,7 @@ python -u infer.py \
 ```
 此外相比于模型训练，预测时还有一些额外的参数，如需要设置 `model_path` 来给出模型所在目录，可以设置 `beam_size` 和 `max_out_len` 来指定 Beam Search 算法的搜索宽度和最大深度（翻译长度），这些参数也可以在 `config.py` 中的 `InferTaskConfig` 内查阅注释说明并进行更改设置。
 
-执行以上预测命令会打印翻译结果到标准输出，每行输出是对应行输入的得分最高的翻译。对于使用 BPE 的英德数据，预测出的翻译结果也将是 BPE 表示的数据，要还原成原始的数据（这里指 tokenize 后的数据）才能进行正确的评估，可以使用以下命令来恢复 `predict.txt` 内的翻译结果到 `predict.tok.txt` 中（无需再次 tokenize 处理）：
+执行以上预测命令会打印翻译结果到 `output_file` 指定的文件中，每行输出是对应行输入的得分最高的翻译。对于使用 BPE 的英德数据，预测出的翻译结果也将是 BPE 表示的数据，要还原成原始的数据（这里指 tokenize 后的数据）才能进行正确的评估，可以使用以下命令来恢复 `predict.txt` 内的翻译结果到 `predict.tok.txt` 中（无需再次 tokenize 处理）：
 ```sh
 sed -r 's/(@@ )|(@@ ?$)//g' predict.txt > predict.tok.txt
 ```
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
index 62e7ce39531c683d3e2a30e51257e0be6e2aa258..959a495d7c597e321182152f7675682297075e83 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/infer.py
@@ -4,9 +4,6 @@ import multiprocessing
 import numpy as np
 import os
 import sys
-if sys.version[0] == '2':
-    reload(sys)
-    sys.setdefaultencoding("utf-8")
 from functools import partial
 
 import paddle
@@ -22,7 +19,7 @@ from train import pad_batch_data, prepare_data_generator
 
 
 def parse_args():
-    parser = argparse.ArgumentParser("Training for Transformer.")
+    parser = argparse.ArgumentParser("Inference for Transformer.")
     parser.add_argument(
         "--src_vocab_fpath",
         type=str,
@@ -38,6 +35,11 @@ def parse_args():
         type=str,
         required=True,
         help="The pattern to match test data files.")
+    parser.add_argument(
+        "--output_file",
+        type=str,
+        default="predict.txt",
+        help="The file to output the translation results of to.")
     parser.add_argument(
         "--batch_size",
         type=int,
@@ -50,14 +52,14 @@ def parse_args():
         help="The buffer size to pool data.")
     parser.add_argument(
         "--special_token",
-        type=lambda x: x.encode(),
-        default=[b"<s>", b"<e>", b"<unk>"],
+        type=lambda x: x.encode("utf8"),
+        default=["<s>", "<e>", "<unk>"],
         nargs=3,
         help="The <bos>, <eos> and <unk> tokens in the dictionary.")
     parser.add_argument(
         "--token_delimiter",
-        type=lambda x: x.encode(),
-        default=b" ",
+        type=lambda x: x.encode("utf8"),
+        default=" ",
         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(
@@ -268,6 +270,7 @@ def fast_infer(args):
     trg_idx2word = reader.DataReader.load_dict(
         dict_path=args.trg_vocab_fpath, reverse=True)
 
+    f = open(args.output_file, "wb")
     while True:
         try:
             feed_dict_list = prepare_feed_dict_list(data_generator, dev_count,
@@ -313,7 +316,7 @@ def fast_infer(args):
                                 np.array(seq_ids)[sub_start:sub_end])
                         ]))
                         scores[i].append(np.array(seq_scores)[sub_end - 1])
-                        print(hyps[i][-1].decode("utf8"))
+                        f.write(hyps[i][-1] + b"\n")
                         if len(hyps[i]) >= InferTaskConfig.n_best:
                             break
         except (StopIteration, fluid.core.EOFException):
@@ -321,6 +324,7 @@ def fast_infer(args):
             if args.use_py_reader:
                 pyreader.reset()
             break
+    f.close()
 
 
 if __name__ == "__main__":
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
index 923e818717d0d90548afb35177a58f461c27ec9c..7e7705fa8990afc86c079f45188bb238c0b5f838 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/reader.py
@@ -182,12 +182,23 @@ class DataReader(object):
                  shuffle=True,
                  shuffle_batch=False,
                  use_token_batch=False,
-                 field_delimiter=b"\t",
-                 token_delimiter=b" ",
-                 start_mark=b"<s>",
-                 end_mark=b"<e>",
-                 unk_mark=b"<unk>",
+                 field_delimiter="\t",
+                 token_delimiter=" ",
+                 start_mark="<s>",
+                 end_mark="<e>",
+                 unk_mark="<unk>",
                  seed=0):
+        # convert str to bytes, and use byte data
+        field_delimiter = field_delimiter if isinstance(
+            field_delimiter, bytes) else field_delimiter.encode("utf8")
+        token_delimiter = token_delimiter if isinstance(
+            token_delimiter, bytes) else token_delimiter.encode("utf8")
+        start_mark = start_mark if isinstance(
+            start_mark, bytes) else start_mark.encode("utf8")
+        end_mark = end_mark if isinstance(end_mark,
+                                          bytes) else end_mark.encode("utf8")
+        unk_mark = unk_mark if isinstance(unk_mark,
+                                          bytes) else unk_mark.encode("utf8")
         self._src_vocab = self.load_dict(src_vocab_fpath)
         self._only_src = True
         if trg_vocab_fpath is not None:
diff --git a/PaddleNLP/unarchived/neural_machine_translation/transformer/train.py b/PaddleNLP/unarchived/neural_machine_translation/transformer/train.py
index 73213de6777455084f1c99d02c014deab1f0bfb4..35681556380bed11e1b0951b7e4026fedcac65c3 100644
--- a/PaddleNLP/unarchived/neural_machine_translation/transformer/train.py
+++ b/PaddleNLP/unarchived/neural_machine_translation/transformer/train.py
@@ -6,9 +6,6 @@ import multiprocessing
 import os
 import six
 import sys
-if sys.version[0] == '2':
-    reload(sys)
-    sys.setdefaultencoding("utf-8")
 import time
 
 import numpy as np
@@ -77,14 +74,14 @@ def parse_args():
         help="The flag indicating whether to shuffle the data batches.")
     parser.add_argument(
         "--special_token",
-        type=lambda x: x.encode(),
-        default=[b"<s>", b"<e>", b"<unk>"],
+        type=lambda x: x.encode("utf8"),
+        default=["<s>", "<e>", "<unk>"],
         nargs=3,
         help="The <bos>, <eos> and <unk> tokens in the dictionary.")
     parser.add_argument(
         "--token_delimiter",
-        type=lambda x: x.encode(),
-        default=b" ",
+        type=lambda x: x.encode("utf8"),
+        default=" ",
         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(