diff --git a/.travis.yml b/.travis.yml
index a5559a796351bb10eacd4739f719b054e085160c..347fb321d761c10d9eecb1e5098ef7cc205ac181 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -1,3 +1,4 @@
+group: deprecated-2017Q2
 language: cpp
 cache: ccache
 sudo: required
diff --git a/.travis/unittest.sh b/.travis/unittest.sh
index 23f15094c8e51132cff350e786654e3bab29cee1..ad223eb4a9c1f57896762ad38d0b3fa5de5c496b 100755
--- a/.travis/unittest.sh
+++ b/.travis/unittest.sh
@@ -8,8 +8,8 @@ abort(){
 
 unittest(){
     cd $1 > /dev/null
-    if [ -f "requirements.txt" ]; then
-        pip install -r requirements.txt
+    if [ -f "setup.sh" ]; then
+        sh setup.sh
     fi
     if [ $? != 0 ]; then
         exit 1
diff --git a/deep_speech_2/README.md b/deep_speech_2/README.md
index 23e0b412b59da4ccfea7a4ce4303faec479ff234..0cdb203d21ef5fa854a011f2f0381078cabcb874 100644
--- a/deep_speech_2/README.md
+++ b/deep_speech_2/README.md
@@ -5,7 +5,7 @@
 Please replace `$PADDLE_INSTALL_DIR` with your own paddle installation directory.
 
 ```
-pip install -r requirements.txt
+sh setup.sh
 export LD_LIBRARY_PATH=$PADDLE_INSTALL_DIR/Paddle/third_party/install/warpctc/lib:$LD_LIBRARY_PATH
 ```
 
diff --git a/deep_speech_2/data_utils/data.py b/deep_speech_2/data_utils/data.py
index 424343a48ffa579a8ab465794987f957de36abdb..44af7ffaa999c618a7dcd4884f528ef60e59eefe 100644
--- a/deep_speech_2/data_utils/data.py
+++ b/deep_speech_2/data_utils/data.py
@@ -7,6 +7,7 @@ from __future__ import print_function
 
 import random
 import numpy as np
+import multiprocessing
 import paddle.v2 as paddle
 from data_utils import utils
 from data_utils.augmentor.augmentation import AugmentationPipeline
@@ -44,6 +45,8 @@ class DataGenerator(object):
     :types max_freq: None|float
     :param specgram_type: Specgram feature type. Options: 'linear'.
     :type specgram_type: str
+    :param num_threads: Number of CPU threads for processing data.
+    :type num_threads: int
     :param random_seed: Random seed.
     :type random_seed: int
     """
@@ -58,6 +61,7 @@ class DataGenerator(object):
                  window_ms=20.0,
                  max_freq=None,
                  specgram_type='linear',
+                 num_threads=multiprocessing.cpu_count(),
                  random_seed=0):
         self._max_duration = max_duration
         self._min_duration = min_duration
@@ -70,6 +74,7 @@ class DataGenerator(object):
             stride_ms=stride_ms,
             window_ms=window_ms,
             max_freq=max_freq)
+        self._num_threads = num_threads
         self._rng = random.Random(random_seed)
         self._epoch = 0
 
@@ -207,10 +212,14 @@ class DataGenerator(object):
 
         def reader():
             for instance in manifest:
-                yield self._process_utterance(instance["audio_filepath"],
-                                              instance["text"])
+                yield instance
 
-        return reader
+        def mapper(instance):
+            return self._process_utterance(instance["audio_filepath"],
+                                           instance["text"])
+
+        return paddle.reader.xmap_readers(
+            mapper, reader, self._num_threads, 1024, order=True)
 
     def _padding_batch(self, batch, padding_to=-1, flatten=False):
         """
diff --git a/deep_speech_2/data_utils/speech.py b/deep_speech_2/data_utils/speech.py
index fc031ff46f4a3820e3b13f7804c91b33948712d1..568e4443ba557149505dfb4de6f230b4962e332a 100644
--- a/deep_speech_2/data_utils/speech.py
+++ b/deep_speech_2/data_utils/speech.py
@@ -94,7 +94,7 @@ class SpeechSegment(AudioSegment):
         return cls(samples, sample_rate, transcripts)
 
     @classmethod
-    def slice_from_file(cls, filepath, start=None, end=None, transcript):
+    def slice_from_file(cls, filepath, transcript, start=None, end=None):
         """Loads a small section of an speech without having to load
         the entire file into the memory which can be incredibly wasteful.
 
diff --git a/deep_speech_2/infer.py b/deep_speech_2/infer.py
index 06449ab05c7960ec78acc9ce5bb664cf1058a845..71518133a347c459bbcf2670fa5d1dc226a619c8 100644
--- a/deep_speech_2/infer.py
+++ b/deep_speech_2/infer.py
@@ -6,6 +6,7 @@ from __future__ import print_function
 import argparse
 import gzip
 import distutils.util
+import multiprocessing
 import paddle.v2 as paddle
 from data_utils.data import DataGenerator
 from model import deep_speech2
@@ -38,6 +39,11 @@ parser.add_argument(
     default=True,
     type=distutils.util.strtobool,
     help="Use gpu or not. (default: %(default)s)")
+parser.add_argument(
+    "--num_threads_data",
+    default=multiprocessing.cpu_count(),
+    type=int,
+    help="Number of cpu threads for preprocessing data. (default: %(default)s)")
 parser.add_argument(
     "--mean_std_filepath",
     default='mean_std.npz',
@@ -67,7 +73,8 @@ def infer():
     data_generator = DataGenerator(
         vocab_filepath=args.vocab_filepath,
         mean_std_filepath=args.mean_std_filepath,
-        augmentation_config='{}')
+        augmentation_config='{}',
+        num_threads=args.num_threads_data)
 
     # create network config
     # paddle.data_type.dense_array is used for variable batch input.
diff --git a/deep_speech_2/requirements.txt b/deep_speech_2/requirements.txt
index c37e88ffe75dfda401726f485fd9928cbb477fab..0183ecf01fc1eb2507e00b3b97a6db92b6e2258e 100644
--- a/deep_speech_2/requirements.txt
+++ b/deep_speech_2/requirements.txt
@@ -1,4 +1,3 @@
 SoundFile==0.9.0.post1
 wget==3.2
-scikits.samplerate==0.3.3
-scipy==0.13.0b1
+scipy==0.13.1
diff --git a/deep_speech_2/setup.sh b/deep_speech_2/setup.sh
new file mode 100644
index 0000000000000000000000000000000000000000..1ae2a5eee0f9cfd5b4318b29cf037165f78f2b73
--- /dev/null
+++ b/deep_speech_2/setup.sh
@@ -0,0 +1,30 @@
+#!/bin/bash
+
+# install python dependencies
+if [ -f 'requirements.txt' ]; then
+    pip install -r requirements.txt
+fi
+if [ $? != 0 ]; then
+    echo "Install python dependencies failed !!!"
+    exit 1
+fi
+
+# install scikits.samplerate
+curl -O "http://www.mega-nerd.com/SRC/libsamplerate-0.1.9.tar.gz"
+if [ $? != 0 ]; then
+    echo "Download libsamplerate-0.1.9.tar.gz failed !!!"
+    exit 1
+fi
+tar -xvf libsamplerate-0.1.9.tar.gz
+cd libsamplerate-0.1.9
+./configure && make && make install
+cd -
+rm -rf libsamplerate-0.1.9
+rm libsamplerate-0.1.9.tar.gz
+pip install scikits.samplerate==0.3.3
+if [ $? != 0 ]; then
+    echo "Install scikits.samplerate failed !!!"
+    exit 1
+fi
+
+echo "Install all dependencies successfully."
diff --git a/deep_speech_2/train.py b/deep_speech_2/train.py
index c60a039b69d91a89eb20e83ec1e090c8600d47a3..fc23ec72692f319b556a75004a7508990df5357e 100644
--- a/deep_speech_2/train.py
+++ b/deep_speech_2/train.py
@@ -9,6 +9,7 @@ import argparse
 import gzip
 import time
 import distutils.util
+import multiprocessing
 import paddle.v2 as paddle
 from model import deep_speech2
 from data_utils.data import DataGenerator
@@ -52,6 +53,18 @@ parser.add_argument(
     default=True,
     type=distutils.util.strtobool,
     help="Use sortagrad or not. (default: %(default)s)")
+parser.add_argument(
+    "--max_duration",
+    default=100.0,
+    type=float,
+    help="Audios with duration larger than this will be discarded. "
+    "(default: %(default)s)")
+parser.add_argument(
+    "--min_duration",
+    default=0.0,
+    type=float,
+    help="Audios with duration smaller than this will be discarded. "
+    "(default: %(default)s)")
 parser.add_argument(
     "--shuffle_method",
     default='instance_shuffle',
@@ -63,6 +76,11 @@ parser.add_argument(
     default=4,
     type=int,
     help="Trainer number. (default: %(default)s)")
+parser.add_argument(
+    "--num_threads_data",
+    default=multiprocessing.cpu_count(),
+    type=int,
+    help="Number of cpu threads for preprocessing data. (default: %(default)s)")
 parser.add_argument(
     "--mean_std_filepath",
     default='mean_std.npz',
@@ -107,7 +125,10 @@ def train():
         return DataGenerator(
             vocab_filepath=args.vocab_filepath,
             mean_std_filepath=args.mean_std_filepath,
-            augmentation_config=args.augmentation_config)
+            augmentation_config=args.augmentation_config,
+            max_duration=args.max_duration,
+            min_duration=args.min_duration,
+            num_threads=args.num_threads_data)
 
     train_generator = data_generator()
     test_generator = data_generator()
diff --git a/language_model/network_conf.py b/language_model/network_conf.py
index 1135ab64087eeafa1675702ba83523b4ad403a3c..e53ca66cc1adc49c6e7dcb50b77219030e74681d 100644
--- a/language_model/network_conf.py
+++ b/language_model/network_conf.py
@@ -51,56 +51,41 @@ def rnn_lm(vocab_size, emb_dim, rnn_type, hidden_size, num_layer):
     return cost, output
 
 
-def ngram_lm(vocab_size, emb_dim, hidden_size, num_layer):
+def ngram_lm(vocab_size, emb_dim, hidden_size, num_layer, gram_num=4):
     """
     N-Gram language model definition.
 
     :param vocab_size: size of vocab.
     :param emb_dim: embedding vector's dimension.
     :param hidden_size: size of unit.
-    :param num_layer: layer number.
+    :param num_layer: number of hidden layers.
+    :param gram_size: gram number in n-gram method
     :return: cost and output layer of model.
     """
 
     assert emb_dim > 0 and hidden_size > 0 and vocab_size > 0 and num_layer > 0
 
-    def wordemb(inlayer):
-        wordemb = paddle.layer.table_projection(
-            input=inlayer,
-            size=emb_dim,
-            param_attr=paddle.attr.Param(
-                name="_proj", initial_std=0.001, learning_rate=1, l2_rate=0))
-        return wordemb
-
     # input layers
-    first_word = paddle.layer.data(
-        name="first_word", type=paddle.data_type.integer_value(vocab_size))
-    second_word = paddle.layer.data(
-        name="second_word", type=paddle.data_type.integer_value(vocab_size))
-    third_word = paddle.layer.data(
-        name="third_word", type=paddle.data_type.integer_value(vocab_size))
-    fourth_word = paddle.layer.data(
-        name="fourth_word", type=paddle.data_type.integer_value(vocab_size))
+    emb_layers = []
+    for i in range(gram_num):
+        word = paddle.layer.data(
+            name="__word%02d__" % (i + 1),
+            type=paddle.data_type.integer_value(vocab_size))
+        emb = paddle.layer.embedding(
+            input=word,
+            size=emb_dim,
+            param_attr=paddle.attr.Param(name="_proj", initial_std=1e-3))
+        emb_layers.append(emb)
     next_word = paddle.layer.data(
-        name="next_word", type=paddle.data_type.integer_value(vocab_size))
-
-    # embedding layer
-    first_emb = wordemb(first_word)
-    second_emb = wordemb(second_word)
-    third_emb = wordemb(third_word)
-    fourth_emb = wordemb(fourth_word)
-
-    context_emb = paddle.layer.concat(
-        input=[first_emb, second_emb, third_emb, fourth_emb])
+        name="__next_word__", type=paddle.data_type.integer_value(vocab_size))
 
     # hidden layer
-    hidden = paddle.layer.fc(
-        input=context_emb, size=hidden_size, act=paddle.activation.Relu())
-    for _ in range(num_layer - 1):
+    for i in range(num_layer):
         hidden = paddle.layer.fc(
-            input=hidden, size=hidden_size, act=paddle.activation.Relu())
+            input=hidden if i else paddle.layer.concat(input=emb_layers),
+            size=hidden_size,
+            act=paddle.activation.Relu())
 
-    # fc(full connected) and output layer
     predict_word = paddle.layer.fc(
         input=[hidden], size=vocab_size, act=paddle.activation.Softmax())
 
diff --git a/nmt_without_attention/README.md b/nmt_without_attention/README.md
index 609126dd872ca15170fa371469019d4057aa8b43..4c63a2589fbc2f4b9649e38a3edc58969409336a 100644
--- a/nmt_without_attention/README.md
+++ b/nmt_without_attention/README.md
@@ -1,8 +1,6 @@
 # 神经网络机器翻译模型
 
 ## 背景介绍
-- PaddleBook中[机器翻译](https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/README.cn.md)的相关章节中，已介绍了带注意力机制（Attention Mechanism）的 Encoder-Decoder 结构，本例则介绍的是不带注意力机制的 Encoder-Decoder 结构。关于注意力机制，读者可进一步参考 PaddleBook 和参考文献\[[3](#参考文献)]。
-
 机器翻译利用计算机将源语言转换成目标语言的同义表达，是自然语言处理中重要的研究方向，有着广泛的应用需求，其实现方式也经历了不断地演化。传统机器翻译方法主要基于规则或统计模型，需要人为地指定翻译规则或设计语言特征，效果依赖于人对源语言与目标语言的理解程度。近些年来，深度学习的提出与迅速发展使得特征的自动学习成为可能。深度学习首先在图像识别和语音识别中取得成功，进而在机器翻译等自然语言处理领域中掀起了研究热潮。机器翻译中的深度学习模型直接学习源语言到目标语言的映射，大为减少了学习过程中人的介入，同时显著地提高了翻译质量。本例介绍在PaddlePaddle中如何利用循环神经网络（Recurrent Neural Network, RNN）构建一个端到端（End-to-End）的神经网络机器翻译（Neural Machine Translation, NMT）模型。
 
 ## 模型概览
@@ -53,14 +51,15 @@ RNN 的原始结构用一个向量来存储隐状态，然而这种结构的 RNN
 在 PaddlePaddle 中，双向编码器可以很方便地调用相关 APIs 实现：
 
 ```python
-#### Encoder
 src_word_id = paddle.layer.data(
     name='source_language_word',
     type=paddle.data_type.integer_value_sequence(source_dict_dim))
+
 # source embedding
 src_embedding = paddle.layer.embedding(
     input=src_word_id, size=word_vector_dim)
-# use bidirectional_gru
+
+# # bidierctional GRU as encoder
 encoded_vector = paddle.networks.bidirectional_gru(
     input=src_embedding,
     size=encoder_size,
@@ -86,18 +85,17 @@ encoded_vector = paddle.networks.bidirectional_gru(
 
 
 ### 无注意力机制的解码器
+-PaddleBook中[机器翻译](https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/README.cn.md)的相关章节中，已介绍了带注意力机制（Attention Mechanism）的 Encoder-Decoder 结构，本例则介绍的是不带注意力机制的 Encoder-Decoder 结构。关于注意力机制，读者可进一步参考 PaddleBook 和参考文献\[[3](#参考文献)]。
 
 对于流行的RNN单元，PaddlePaddle 已有很好的实现均可直接调用。如果希望在 RNN 每一个时间步实现某些自定义操作，可使用 PaddlePaddle 中的`recurrent_layer_group`。首先，自定义单步逻辑函数，再利用函数 `recurrent_group()` 循环调用单步逻辑函数处理整个序列。本例中的无注意力机制的解码器便是使用`recurrent_layer_group`来实现，其中，单步逻辑函数`gru_decoder_without_attention()`相关代码如下：
 
 ```python
-#### Decoder
+# the initialization state for decoder GRU
 encoder_last = paddle.layer.last_seq(input=encoded_vector)
-encoder_last_projected = paddle.layer.mixed(
-        size=decoder_size,
-        act=paddle.activation.Tanh(),
-        input=paddle.layer.full_matrix_projection(input=encoder_last))
+encoder_last_projected = paddle.layer.fc(
+    size=decoder_size, act=paddle.activation.Tanh(), input=encoder_last)
 
-# gru step
+# the step function for decoder GRU
 def gru_decoder_without_attention(enc_vec, current_word):
     '''
     Step function for gru decoder
@@ -107,33 +105,29 @@ def gru_decoder_without_attention(enc_vec, current_word):
     :type current_word: layer object
     '''
     decoder_mem = paddle.layer.memory(
-        name='gru_decoder',
-        size=decoder_size,
-        boot_layer=encoder_last_projected)
+            name="gru_decoder",
+            size=decoder_size,
+            boot_layer=encoder_last_projected)
 
     context = paddle.layer.last_seq(input=enc_vec)
 
-    decoder_inputs = paddle.layer.mixed(
-        size=decoder_size * 3,
-        input=[
-            paddle.layer.full_matrix_projection(input=context),
-            paddle.layer.full_matrix_projection(input=current_word)
-        ])
+    decoder_inputs = paddle.layer.fc(
+        size=decoder_size * 3, input=[context, current_word])
 
     gru_step = paddle.layer.gru_step(
-        name='gru_decoder',
+        name="gru_decoder",
         act=paddle.activation.Tanh(),
         gate_act=paddle.activation.Sigmoid(),
         input=decoder_inputs,
         output_mem=decoder_mem,
         size=decoder_size)
 
-    out = paddle.layer.mixed(
+     out = paddle.layer.fc(
         size=target_dict_dim,
         bias_attr=True,
         act=paddle.activation.Softmax(),
-        input=paddle.layer.full_matrix_projection(input=gru_step))
-    return out
+        input=gru_step)
+    return out  
 ```
 
 在模型训练和测试阶段，解码器的行为有很大的不同：
@@ -144,34 +138,14 @@ def gru_decoder_without_attention(enc_vec, current_word):
 训练和生成的逻辑分别实现在如下的`if-else`条件分支中：
 
 ```python
-decoder_group_name = "decoder_group"
-group_input1 = paddle.layer.StaticInput(input=encoded_vector, is_seq=True)
+group_input1 = paddle.layer.StaticInput(input=encoded_vector)
 group_inputs = [group_input1]
-if not generating:
-    trg_embedding = paddle.layer.embedding(
-        input=paddle.layer.data(
-            name='target_language_word',
-            type=paddle.data_type.integer_value_sequence(target_dict_dim)),
-            size=word_vector_dim,
-            param_attr=paddle.attr.ParamAttr(name='_target_language_embedding'))
-    group_inputs.append(trg_embedding)
-
-    decoder = paddle.layer.recurrent_group(
-        name=decoder_group_name,
-        step=gru_decoder_without_attention,
-        input=group_inputs)
-
-    lbl = paddle.layer.data(
-        name='target_language_next_word',
-            type=paddle.data_type.integer_value_sequence(target_dict_dim))
-    cost = paddle.layer.classification_cost(input=decoder, label=lbl)
-
-    return cost
-else:
 
+decoder_group_name = "decoder_group"
+if is_generating:
     trg_embedding = paddle.layer.GeneratedInput(
         size=target_dict_dim,
-        embedding_name='_target_language_embedding',
+        embedding_name="_target_language_embedding",
         embedding_size=word_vector_dim)
     group_inputs.append(trg_embedding)
 
@@ -185,6 +159,26 @@ else:
         max_length=max_length)
 
     return beam_gen
+else:
+    trg_embedding = paddle.layer.embedding(
+        input=paddle.layer.data(
+            name="target_language_word",
+            type=paddle.data_type.integer_value_sequence(target_dict_dim)),
+        size=word_vector_dim,
+        param_attr=paddle.attr.ParamAttr(name="_target_language_embedding"))
+    group_inputs.append(trg_embedding)
+
+    decoder = paddle.layer.recurrent_group(
+        name=decoder_group_name,
+        step=gru_decoder_without_attention,
+        input=group_inputs)
+
+    lbl = paddle.layer.data(
+        name="target_language_next_word",
+        type=paddle.data_type.integer_value_sequence(target_dict_dim))
+    cost = paddle.layer.classification_cost(input=decoder, label=lbl)
+
+    return cost
 ```
 
 ## 数据准备
@@ -208,13 +202,16 @@ parameters = paddle.parameters.create(cost)
 **b) 设定训练过程中的优化策略、定义训练数据读取 `reader`**
 
 ```python
-# define optimize method and trainer
+# define optimization method
 optimizer = paddle.optimizer.RMSProp(
     learning_rate=1e-3,
     gradient_clipping_threshold=10.0,
     regularization=paddle.optimizer.L2Regularization(rate=8e-4))
+
+# define the trainer instance
 trainer = paddle.trainer.SGD(
     cost=cost, parameters=parameters, update_equation=optimizer)
+
 # define data reader
 wmt14_reader = paddle.batch(
     paddle.reader.shuffle(
@@ -225,20 +222,19 @@ wmt14_reader = paddle.batch(
 **c) 定义事件句柄，打印训练中间结果、保存模型快照**
 
 ```python
-# define event_handler callback
+# define the event_handler callback
 def event_handler(event):
     if isinstance(event, paddle.event.EndIteration):
-        if event.batch_id % 100 == 0 and event.batch_id > 0:
-            with gzip.open('models/nmt_without_att_params_batch_%d.tar.gz' %
-                           event.batch_id, 'w') as f:
+        if not event.batch_id % 100 and event.batch_id:
+            with gzip.open(
+                    os.path.join(save_path,
+                                 "nmt_without_att_%05d_batch_%05d.tar.gz" %
+                                 event.pass_id, event.batch_id), "w") as f:
                 parameters.to_tar(f)
 
-        if event.batch_id % 10 == 0:
-            print "\nPass %d, Batch %d, Cost%f, %s" % (
-                event.pass_id, event.batch_id, event.cost, event.metrics)
-        else:
-            sys.stdout.write('.')
-            sys.stdout.flush()
+        if event.batch_id and not event.batch_id % 10:
+            logger.info("Pass %d, Batch %d, Cost %f, %s" % (
+                event.pass_id, event.batch_id, event.cost, event.metrics))
 ```
 
 **d) 开始训练**
@@ -300,26 +296,22 @@ beam_result = paddle.infer(
 **c) 加载源语言和目标语言词典，将`id`序列表示的句子转化成原语言并输出结果**
 
 ```python
-# get the dictionary
-src_dict, trg_dict = paddle.dataset.wmt14.get_dict(source_dict_dim)
-
-# the delimited element of generated sequences is -1,
-# the first element of each generated sequence is the sequence length
-seq_list = []
-seq = []
-for w in beam_result[1]:
-    if w != -1:
-        seq.append(w)
-    else:
-        seq_list.append(' '.join([trg_dict.get(w) for w in seq[1:]]))
-        seq = []
-
-prob = beam_result[0]
-for i in xrange(len(gen_data)):
-    print "\n*******************************************************\n"
-    print "src:", ' '.join([src_dict.get(w) for w in gen_data[i][0]]), "\n"
+beam_result = inferer.infer(input=test_batch, field=["prob", "id"])
+
+gen_sen_idx = np.where(beam_result[1] == -1)[0]
+assert len(gen_sen_idx) == len(test_batch) * beam_size
+
+start_pos, end_pos = 1, 0
+for i, sample in enumerate(test_batch):
+    print(" ".join([
+        src_dict[w] for w in sample[0][1:-1]
+    ]))  # skip the start and ending mark when print the source sentence
     for j in xrange(beam_size):
-        print "prob = %f:" % (prob[i][j]), seq_list[i * beam_size + j]
+        end_pos = gen_sen_idx[i * beam_size + j]
+        print("%.4f\t%s" % (beam_result[0][i][j], " ".join(
+            trg_dict[w] for w in beam_result[1][start_pos:end_pos])))
+        start_pos = end_pos + 2
+    print("\n")
 ```
 
 模型测试的执行与模型训练类似，只需执行
@@ -327,23 +319,20 @@ for i in xrange(len(gen_data)):
 ```bash
 python generate.py
 ```
-则自动为测试数据生成了对应的翻译结果。
-设置beam search的宽度为3，输入某个法文句子
 
-```text
-src: <s> Elles connaissent leur entreprise mieux que personne . <e>
-```
-
-其对应的英文翻译结果为
+设置beam search的宽度为3，输入为一个法文句子，则自动为测试数据生成对应的翻译结果，输出格式如下：
 
 ```text
-prob = -3.754819: They know their business better than anyone . <e>
-prob = -4.445528: They know their businesses better than anyone . <e>
-prob = -5.026885: They know their business better than anybody . <e>
-```
+Elles connaissent leur entreprise mieux que personne .
+-3.754819        They know their business better than anyone . <e>
+-4.445528        They know their businesses better than anyone . <e>
+-5.026885        They know their business better than anybody . <e>
 
-* `prob`表示生成句子的得分，随之其后则是翻译生成的句子；
-* `<s>` 表示句子的开始，`<e>`表示一个句子的结束，如果出现了在词典中未包含的词，则用`<unk>`替代。
+```
+- 第一行为输入的源语言句子。
+- 第二 ~ `beam_size + 1` 行是柱搜索生成的 `beam_size` 条翻译结果
+    - 一行之内以“\t”分隔为两列，第一列是句子的log 概率，第二列是翻译结果的文本。
+    - `<s>` 表示句子的开始，`<e>`表示一个句子的结束，如果出现了在词典中未包含的词，则用`<unk>`替代。
 
 至此，我们在 PaddlePaddle 上实现了一个初步的机器翻译模型。我们可以看到，PaddlePaddle 提供了灵活丰富的API供大家选择和使用，使得我们能够很方便完成各种复杂网络的配置。机器翻译本身也是个快速发展的领域，各种新方法新思想在不断涌现。在学习完本例后，读者若有兴趣和余力，可基于 PaddlePaddle 平台实现更为复杂、性能更优的机器翻译模型。
 
diff --git a/nmt_without_attention/generate.py b/nmt_without_attention/generate.py
index e3c220685e2bc75a13f57d4f38faa28cb31a4f65..074db9e93ee459b8b37b55aed47af6473750aca0 100644
--- a/nmt_without_attention/generate.py
+++ b/nmt_without_attention/generate.py
@@ -1,32 +1,37 @@
 #!/usr/bin/env python
 import os
-from network_conf import *
+import logging
+import numpy as np
+
+from network_conf import seq2seq_net
+
+logger = logging.getLogger("paddle")
+logger.setLevel(logging.WARNING)
 
 
 def infer_a_batch(inferer, test_batch, beam_size, src_dict, trg_dict):
     beam_result = inferer.infer(input=test_batch, field=["prob", "id"])
 
-    # the delimited element of generated sequences is -1,
-    # the first element of each generated sequence is the sequence length
-    seq_list, seq = [], []
-    for w in beam_result[1]:
-        if w != -1:
-            seq.append(w)
-        else:
-            seq_list.append(" ".join([trg_dict.get(w) for w in seq[1:]]))
-            seq = []
-
-    prob = beam_result[0]
+    gen_sen_idx = np.where(beam_result[1] == -1)[0]
+    assert len(gen_sen_idx) == len(test_batch) * beam_size
+
+    start_pos, end_pos = 1, 0
     for i, sample in enumerate(test_batch):
-        print("src:", " ".join([src_dict.get(w) for w in sample[0]]), "\n")
+        print(" ".join([
+            src_dict[w] for w in sample[0][1:-1]
+        ]))  # skip the start and ending mark when print the source sentence
         for j in xrange(beam_size):
-            print("prob = %f:" % (prob[i][j]), seq_list[i * beam_size + j])
+            end_pos = gen_sen_idx[i * beam_size + j]
+            print("%.4f\t%s" % (beam_result[0][i][j], " ".join(
+                trg_dict[w] for w in beam_result[1][start_pos:end_pos])))
+            start_pos = end_pos + 2
         print("\n")
 
 
-def generate(source_dict_dim, target_dict_dim, model_path, batch_size):
+def generate(source_dict_dim, target_dict_dim, model_path, beam_size,
+             batch_size):
     """
-    Generating function for NMT
+    sequence generation for NMT
 
     :param source_dict_dim: size of source dictionary
     :type source_dict_dim: int
@@ -34,16 +39,19 @@ def generate(source_dict_dim, target_dict_dim, model_path, batch_size):
     :type target_dict_dim: int
     :param model_path: path for inital model
     :type model_path: string
+    :param beam_size: the expanson width in each generation setp
+    :param beam_size: int
+    :param batch_size: the number of training examples in one forward pass
+    :param batch_size: int
     """
 
     assert os.path.exists(model_path), "trained model does not exist."
 
     # step 1: prepare dictionary
     src_dict, trg_dict = paddle.dataset.wmt14.get_dict(source_dict_dim)
-    beam_size = 5
 
     # step 2: load the trained model
-    paddle.init(use_gpu=True, trainer_count=1)
+    paddle.init(use_gpu=False, trainer_count=1)
     with gzip.open(model_path) as f:
         parameters = paddle.parameters.Parameters.from_tar(f)
     beam_gen = seq2seq_net(
@@ -72,5 +80,6 @@ if __name__ == "__main__":
     generate(
         source_dict_dim=3000,
         target_dict_dim=3000,
-        batch_size=5,
-        model_path="models/nmt_without_att_params_batch_00001.tar.gz")
+        batch_size=20,
+        beam_size=5,
+        model_path="models/nmt_without_att_params_batch_00347.tar.gz")
diff --git a/nmt_without_attention/index.html b/nmt_without_attention/index.html
index b68bdb5b6d1c84f8cde6f66583788f678c79e858..b6ddaac64203423ac4d428b2fd965672aa98baff 100644
--- a/nmt_without_attention/index.html
+++ b/nmt_without_attention/index.html
@@ -43,8 +43,6 @@
 # 神经网络机器翻译模型
 
 ## 背景介绍
-- PaddleBook中[机器翻译](https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/README.cn.md)的相关章节中，已介绍了带注意力机制（Attention Mechanism）的 Encoder-Decoder 结构，本例则介绍的是不带注意力机制的 Encoder-Decoder 结构。关于注意力机制，读者可进一步参考 PaddleBook 和参考文献\[[3](#参考文献)]。
-
 机器翻译利用计算机将源语言转换成目标语言的同义表达，是自然语言处理中重要的研究方向，有着广泛的应用需求，其实现方式也经历了不断地演化。传统机器翻译方法主要基于规则或统计模型，需要人为地指定翻译规则或设计语言特征，效果依赖于人对源语言与目标语言的理解程度。近些年来，深度学习的提出与迅速发展使得特征的自动学习成为可能。深度学习首先在图像识别和语音识别中取得成功，进而在机器翻译等自然语言处理领域中掀起了研究热潮。机器翻译中的深度学习模型直接学习源语言到目标语言的映射，大为减少了学习过程中人的介入，同时显著地提高了翻译质量。本例介绍在PaddlePaddle中如何利用循环神经网络（Recurrent Neural Network, RNN）构建一个端到端（End-to-End）的神经网络机器翻译（Neural Machine Translation, NMT）模型。
 
 ## 模型概览
@@ -95,14 +93,15 @@ RNN 的原始结构用一个向量来存储隐状态，然而这种结构的 RNN
 在 PaddlePaddle 中，双向编码器可以很方便地调用相关 APIs 实现：
 
 ```python
-#### Encoder
 src_word_id = paddle.layer.data(
     name='source_language_word',
     type=paddle.data_type.integer_value_sequence(source_dict_dim))
+
 # source embedding
 src_embedding = paddle.layer.embedding(
     input=src_word_id, size=word_vector_dim)
-# use bidirectional_gru
+
+# # bidierctional GRU as encoder
 encoded_vector = paddle.networks.bidirectional_gru(
     input=src_embedding,
     size=encoder_size,
@@ -128,18 +127,17 @@ encoded_vector = paddle.networks.bidirectional_gru(
 
 
 ### 无注意力机制的解码器
+-PaddleBook中[机器翻译](https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/README.cn.md)的相关章节中，已介绍了带注意力机制（Attention Mechanism）的 Encoder-Decoder 结构，本例则介绍的是不带注意力机制的 Encoder-Decoder 结构。关于注意力机制，读者可进一步参考 PaddleBook 和参考文献\[[3](#参考文献)]。
 
 对于流行的RNN单元，PaddlePaddle 已有很好的实现均可直接调用。如果希望在 RNN 每一个时间步实现某些自定义操作，可使用 PaddlePaddle 中的`recurrent_layer_group`。首先，自定义单步逻辑函数，再利用函数 `recurrent_group()` 循环调用单步逻辑函数处理整个序列。本例中的无注意力机制的解码器便是使用`recurrent_layer_group`来实现，其中，单步逻辑函数`gru_decoder_without_attention()`相关代码如下：
 
 ```python
-#### Decoder
+# the initialization state for decoder GRU
 encoder_last = paddle.layer.last_seq(input=encoded_vector)
-encoder_last_projected = paddle.layer.mixed(
-        size=decoder_size,
-        act=paddle.activation.Tanh(),
-        input=paddle.layer.full_matrix_projection(input=encoder_last))
+encoder_last_projected = paddle.layer.fc(
+    size=decoder_size, act=paddle.activation.Tanh(), input=encoder_last)
 
-# gru step
+# the step function for decoder GRU
 def gru_decoder_without_attention(enc_vec, current_word):
     '''
     Step function for gru decoder
@@ -149,33 +147,29 @@ def gru_decoder_without_attention(enc_vec, current_word):
     :type current_word: layer object
     '''
     decoder_mem = paddle.layer.memory(
-        name='gru_decoder',
-        size=decoder_size,
-        boot_layer=encoder_last_projected)
+            name="gru_decoder",
+            size=decoder_size,
+            boot_layer=encoder_last_projected)
 
     context = paddle.layer.last_seq(input=enc_vec)
 
-    decoder_inputs = paddle.layer.mixed(
-        size=decoder_size * 3,
-        input=[
-            paddle.layer.full_matrix_projection(input=context),
-            paddle.layer.full_matrix_projection(input=current_word)
-        ])
+    decoder_inputs = paddle.layer.fc(
+        size=decoder_size * 3, input=[context, current_word])
 
     gru_step = paddle.layer.gru_step(
-        name='gru_decoder',
+        name="gru_decoder",
         act=paddle.activation.Tanh(),
         gate_act=paddle.activation.Sigmoid(),
         input=decoder_inputs,
         output_mem=decoder_mem,
         size=decoder_size)
 
-    out = paddle.layer.mixed(
+     out = paddle.layer.fc(
         size=target_dict_dim,
         bias_attr=True,
         act=paddle.activation.Softmax(),
-        input=paddle.layer.full_matrix_projection(input=gru_step))
-    return out
+        input=gru_step)
+    return out  
 ```
 
 在模型训练和测试阶段，解码器的行为有很大的不同：
@@ -186,34 +180,14 @@ def gru_decoder_without_attention(enc_vec, current_word):
 训练和生成的逻辑分别实现在如下的`if-else`条件分支中：
 
 ```python
-decoder_group_name = "decoder_group"
-group_input1 = paddle.layer.StaticInput(input=encoded_vector, is_seq=True)
+group_input1 = paddle.layer.StaticInput(input=encoded_vector)
 group_inputs = [group_input1]
-if not generating:
-    trg_embedding = paddle.layer.embedding(
-        input=paddle.layer.data(
-            name='target_language_word',
-            type=paddle.data_type.integer_value_sequence(target_dict_dim)),
-            size=word_vector_dim,
-            param_attr=paddle.attr.ParamAttr(name='_target_language_embedding'))
-    group_inputs.append(trg_embedding)
-
-    decoder = paddle.layer.recurrent_group(
-        name=decoder_group_name,
-        step=gru_decoder_without_attention,
-        input=group_inputs)
-
-    lbl = paddle.layer.data(
-        name='target_language_next_word',
-            type=paddle.data_type.integer_value_sequence(target_dict_dim))
-    cost = paddle.layer.classification_cost(input=decoder, label=lbl)
-
-    return cost
-else:
 
+decoder_group_name = "decoder_group"
+if is_generating:
     trg_embedding = paddle.layer.GeneratedInput(
         size=target_dict_dim,
-        embedding_name='_target_language_embedding',
+        embedding_name="_target_language_embedding",
         embedding_size=word_vector_dim)
     group_inputs.append(trg_embedding)
 
@@ -227,6 +201,26 @@ else:
         max_length=max_length)
 
     return beam_gen
+else:
+    trg_embedding = paddle.layer.embedding(
+        input=paddle.layer.data(
+            name="target_language_word",
+            type=paddle.data_type.integer_value_sequence(target_dict_dim)),
+        size=word_vector_dim,
+        param_attr=paddle.attr.ParamAttr(name="_target_language_embedding"))
+    group_inputs.append(trg_embedding)
+
+    decoder = paddle.layer.recurrent_group(
+        name=decoder_group_name,
+        step=gru_decoder_without_attention,
+        input=group_inputs)
+
+    lbl = paddle.layer.data(
+        name="target_language_next_word",
+        type=paddle.data_type.integer_value_sequence(target_dict_dim))
+    cost = paddle.layer.classification_cost(input=decoder, label=lbl)
+
+    return cost
 ```
 
 ## 数据准备
@@ -250,13 +244,16 @@ parameters = paddle.parameters.create(cost)
 **b) 设定训练过程中的优化策略、定义训练数据读取 `reader`**
 
 ```python
-# define optimize method and trainer
+# define optimization method
 optimizer = paddle.optimizer.RMSProp(
     learning_rate=1e-3,
     gradient_clipping_threshold=10.0,
     regularization=paddle.optimizer.L2Regularization(rate=8e-4))
+
+# define the trainer instance
 trainer = paddle.trainer.SGD(
     cost=cost, parameters=parameters, update_equation=optimizer)
+
 # define data reader
 wmt14_reader = paddle.batch(
     paddle.reader.shuffle(
@@ -267,20 +264,19 @@ wmt14_reader = paddle.batch(
 **c) 定义事件句柄，打印训练中间结果、保存模型快照**
 
 ```python
-# define event_handler callback
+# define the event_handler callback
 def event_handler(event):
     if isinstance(event, paddle.event.EndIteration):
-        if event.batch_id % 100 == 0 and event.batch_id > 0:
-            with gzip.open('models/nmt_without_att_params_batch_%d.tar.gz' %
-                           event.batch_id, 'w') as f:
+        if not event.batch_id % 100 and event.batch_id:
+            with gzip.open(
+                    os.path.join(save_path,
+                                 "nmt_without_att_%05d_batch_%05d.tar.gz" %
+                                 event.pass_id, event.batch_id), "w") as f:
                 parameters.to_tar(f)
 
-        if event.batch_id % 10 == 0:
-            print "\nPass %d, Batch %d, Cost%f, %s" % (
-                event.pass_id, event.batch_id, event.cost, event.metrics)
-        else:
-            sys.stdout.write('.')
-            sys.stdout.flush()
+        if event.batch_id and not event.batch_id % 10:
+            logger.info("Pass %d, Batch %d, Cost %f, %s" % (
+                event.pass_id, event.batch_id, event.cost, event.metrics))
 ```
 
 **d) 开始训练**
@@ -342,26 +338,22 @@ beam_result = paddle.infer(
 **c) 加载源语言和目标语言词典，将`id`序列表示的句子转化成原语言并输出结果**
 
 ```python
-# get the dictionary
-src_dict, trg_dict = paddle.dataset.wmt14.get_dict(source_dict_dim)
-
-# the delimited element of generated sequences is -1,
-# the first element of each generated sequence is the sequence length
-seq_list = []
-seq = []
-for w in beam_result[1]:
-    if w != -1:
-        seq.append(w)
-    else:
-        seq_list.append(' '.join([trg_dict.get(w) for w in seq[1:]]))
-        seq = []
-
-prob = beam_result[0]
-for i in xrange(len(gen_data)):
-    print "\n*******************************************************\n"
-    print "src:", ' '.join([src_dict.get(w) for w in gen_data[i][0]]), "\n"
+beam_result = inferer.infer(input=test_batch, field=["prob", "id"])
+
+gen_sen_idx = np.where(beam_result[1] == -1)[0]
+assert len(gen_sen_idx) == len(test_batch) * beam_size
+
+start_pos, end_pos = 1, 0
+for i, sample in enumerate(test_batch):
+    print(" ".join([
+        src_dict[w] for w in sample[0][1:-1]
+    ]))  # skip the start and ending mark when print the source sentence
     for j in xrange(beam_size):
-        print "prob = %f:" % (prob[i][j]), seq_list[i * beam_size + j]
+        end_pos = gen_sen_idx[i * beam_size + j]
+        print("%.4f\t%s" % (beam_result[0][i][j], " ".join(
+            trg_dict[w] for w in beam_result[1][start_pos:end_pos])))
+        start_pos = end_pos + 2
+    print("\n")
 ```
 
 模型测试的执行与模型训练类似，只需执行
@@ -369,23 +361,20 @@ for i in xrange(len(gen_data)):
 ```bash
 python generate.py
 ```
-则自动为测试数据生成了对应的翻译结果。
-设置beam search的宽度为3，输入某个法文句子
 
-```text
-src: <s> Elles connaissent leur entreprise mieux que personne . <e>
-```
-
-其对应的英文翻译结果为
+设置beam search的宽度为3，输入为一个法文句子，则自动为测试数据生成对应的翻译结果，输出格式如下：
 
 ```text
-prob = -3.754819: They know their business better than anyone . <e>
-prob = -4.445528: They know their businesses better than anyone . <e>
-prob = -5.026885: They know their business better than anybody . <e>
-```
+Elles connaissent leur entreprise mieux que personne .
+-3.754819        They know their business better than anyone . <e>
+-4.445528        They know their businesses better than anyone . <e>
+-5.026885        They know their business better than anybody . <e>
 
-* `prob`表示生成句子的得分，随之其后则是翻译生成的句子；
-* `<s>` 表示句子的开始，`<e>`表示一个句子的结束，如果出现了在词典中未包含的词，则用`<unk>`替代。
+```
+- 第一行为输入的源语言句子。
+- 第二 ~ `beam_size + 1` 行是柱搜索生成的 `beam_size` 条翻译结果
+    - 一行之内以“\t”分隔为两列，第一列是句子的log 概率，第二列是翻译结果的文本。
+    - `<s>` 表示句子的开始，`<e>`表示一个句子的结束，如果出现了在词典中未包含的词，则用`<unk>`替代。
 
 至此，我们在 PaddlePaddle 上实现了一个初步的机器翻译模型。我们可以看到，PaddlePaddle 提供了灵活丰富的API供大家选择和使用，使得我们能够很方便完成各种复杂网络的配置。机器翻译本身也是个快速发展的领域，各种新方法新思想在不断涌现。在学习完本例后，读者若有兴趣和余力，可基于 PaddlePaddle 平台实现更为复杂、性能更优的机器翻译模型。
 
diff --git a/nmt_without_attention/train.py b/nmt_without_attention/train.py
index 6b847fa8e04334407f127fb3a5ed78f610111fb1..9600df8e5b70cca90543062b040e6dddc540440c 100644
--- a/nmt_without_attention/train.py
+++ b/nmt_without_attention/train.py
@@ -1,56 +1,66 @@
 #!/usr/bin/env python
+import os
+import logging
+import paddle.v2 as paddle
 
-from network_conf import *
+from network_conf import seq2seq_net
 
+logger = logging.getLogger("paddle")
+logger.setLevel(logging.INFO)
 
-def train(source_dict_dim, target_dict_dim):
+
+def train(save_dir_path, source_dict_dim, target_dict_dim):
     '''
     Training function for NMT
 
+    :param save_dir_path: path of the directory to save the trained models.
+    :param save_dir_path: str
     :param source_dict_dim: size of source dictionary
     :type source_dict_dim: int
     :param target_dict_dim: size of target dictionary
     :type target_dict_dim: int
     '''
-    # initialize model
+    if not os.path.exists(save_dir_path):
+        os.mkdir(save_dir_path)
+
+    # initialize PaddlePaddle
     paddle.init(use_gpu=False, trainer_count=1)
 
     cost = seq2seq_net(source_dict_dim, target_dict_dim)
     parameters = paddle.parameters.create(cost)
 
-    # define optimize method and trainer
+    # define optimization method and the trainer instance
     optimizer = paddle.optimizer.RMSProp(
         learning_rate=1e-3,
         gradient_clipping_threshold=10.0,
         regularization=paddle.optimizer.L2Regularization(rate=8e-4))
     trainer = paddle.trainer.SGD(
         cost=cost, parameters=parameters, update_equation=optimizer)
+
     # define data reader
     wmt14_reader = paddle.batch(
         paddle.reader.shuffle(
             paddle.dataset.wmt14.train(source_dict_dim), buf_size=8192),
         batch_size=8)
 
-    # define event_handler callback
+    # define the event_handler callback
     def event_handler(event):
         if isinstance(event, paddle.event.EndIteration):
-            if not event.batch_id % 500 and event.batch_id:
-                with gzip.open("models/nmt_without_att_params_batch_%05d.tar.gz"
-                               % event.batch_id, "w") as f:
+            if not event.batch_id % 100 and event.batch_id:
+                with gzip.open(
+                        os.path.join(save_path,
+                                     "nmt_without_att_%05d_batch_%05d.tar.gz" %
+                                     event.pass_id, event.batch_id), "w") as f:
                     parameters.to_tar(f)
 
             if event.batch_id and not event.batch_id % 10:
-                print("\nPass %d, Batch %d, Cost %f, %s" %
-                      (event.pass_id, event.batch_id, event.cost,
-                       event.metrics))
-            else:
-                sys.stdout.write('.')
-                sys.stdout.flush()
-
-    # start to train
+                logger.info("Pass %d, Batch %d, Cost %f, %s" % (
+                    event.pass_id, event.batch_id, event.cost, event.metrics))
+
+    # start training
     trainer.train(
         reader=wmt14_reader, event_handler=event_handler, num_passes=2)
 
 
 if __name__ == '__main__':
-    train(source_dict_dim=3000, target_dict_dim=3000)
+    train(save_dir_path="models", source_dict_dim=3000, target_dict_dim=3000)