Merge branch 'develop' of http://github.com/paddlepaddle/paddle into patch-1

.gitignore

+paddle/operators/check_t.save
+paddle/operators/check_tensor.ls
+paddle/operators/tensor.save
+python/paddle/v2/fluid/tests/book/image_classification_resnet.inference.model/
+python/paddle/v2/fluid/tests/book/image_classification_vgg.inference.model/
+python/paddle/v2/fluid/tests/book/label_semantic_roles.inference.model/
 *.DS_Store
 build/
 build_doc/
@@ -27,5 +33,5 @@ CMakeFiles
 cmake_install.cmake
 paddle/.timestamp
 python/paddlepaddle.egg-info/
-paddle/pybind/pybind.h
+paddle/fluid/pybind/pybind.h
 python/paddle/version.py

CMakeLists.txt

@@ -137,7 +137,7 @@ include(external/openblas)  # download, build, install openblas
 include(external/mkldnn)    # download, build, install mkldnn
 include(external/swig)      # download, build, install swig
 include(external/warpctc)   # download, build, install warpctc
-include(external/boost)     # download, build, install boost
+include(external/boost)     # download boost
 include(external/any)       # download libn::any
 include(external/eigen)     # download eigen3
 include(external/pybind11)  # download pybind11
@@ -156,6 +156,7 @@ include(rdma)               # set rdma libraries
 include(flags)              # set paddle compile flags
 include(version)            # set PADDLE_VERSION
 include(coveralls)          # set code coverage
+include(inference_lib)      # add paddle fluid inference libraries
 
 
 include_directories("${PADDLE_SOURCE_DIR}")

CONTRIBUTING.md

 # Contribute Code
 
+You are welcome to contribute to project PaddlePaddle. To contribute to PaddlePaddle, you have to agree with the 
+[PaddlePaddle Contributor License Agreement](https://gist.github.com/wangkuiyi/0c22c7b1bd3bb7eb27d76f85c3a3e329).
+
 We sincerely appreciate your contribution.  This document explains our workflow and work style.
 
 ## Workflow

cmake/cuda.cmake

@@ -181,7 +181,8 @@ elseif(CMAKE_BUILD_TYPE  STREQUAL "Release")
 elseif(CMAKE_BUILD_TYPE  STREQUAL "RelWithDebInfo")
     list(APPEND CUDA_NVCC_FLAGS  ${CMAKE_CXX_FLAGS_RELWITHDEBINFO})
 elseif(CMAKE_BUILD_TYPE  STREQUAL "MinSizeRel")
-    list(APPEND CUDA_NVCC_FLAGS  ${CMAKE_CXX_FLAGS_MINSIZEREL})
+    # nvcc 9 does not support -Os. Use Release flags instead
+    list(APPEND CUDA_NVCC_FLAGS  ${CMAKE_CXX_FLAGS_RELEASE})
 endif()
 
 mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD)

cmake/external/boost.cmake

@@ -21,6 +21,7 @@ set(BOOST_URL           "http://sourceforge.net/projects/boost/files/boost/${BOO
 set(BOOST_SOURCES_DIR ${THIRD_PARTY_PATH}/boost)
 set(BOOST_DOWNLOAD_DIR  "${BOOST_SOURCES_DIR}/src/${BOOST_PROJECT}")
 set(BOOST_INCLUDE_DIR "${BOOST_DOWNLOAD_DIR}/${BOOST_TAR}" CACHE PATH "boost include directory." FORCE)
+set_directory_properties(PROPERTIES CLEAN_NO_CUSTOM 1)
 
 include_directories(${BOOST_INCLUDE_DIR})
 

cmake/external/eigen.cmake

@@ -28,9 +28,3 @@ endif()
 add_dependencies(eigen3 extern_eigen3)
 
 LIST(APPEND external_project_dependencies eigen3)
-
-IF(NOT WITH_C_API AND WITH_FLUID)
-    INSTALL(FILES ${EIGEN_INCLUDE_DIR}/Eigen/Core DESTINATION third_party/eigen3/Eigen)
-    INSTALL(DIRECTORY ${EIGEN_INCLUDE_DIR}/Eigen/src DESTINATION third_party/eigen3/Eigen)
-    INSTALL(DIRECTORY ${EIGEN_INCLUDE_DIR}/unsupported/Eigen DESTINATION third_party/eigen3/unsupported)
-ENDIF()

cmake/external/gflags.cmake

@@ -52,7 +52,7 @@ ADD_DEPENDENCIES(gflags extern_gflags)
 
 LIST(APPEND external_project_dependencies gflags)
 
-IF(WITH_C_API OR WITH_FLUID)
+IF(WITH_C_API)
   INSTALL(DIRECTORY ${GFLAGS_INCLUDE_DIR} DESTINATION third_party/gflags)
   IF(ANDROID)
     INSTALL(FILES ${GFLAGS_LIBRARIES} DESTINATION third_party/gflags/lib/${ANDROID_ABI})

cmake/external/glog.cmake

@@ -68,7 +68,7 @@ LINK_LIBRARIES(glog gflags)
 
 LIST(APPEND external_project_dependencies glog)
 
-IF(WITH_C_API OR WITH_FLUID)
+IF(WITH_C_API)
   INSTALL(DIRECTORY ${GLOG_INCLUDE_DIR} DESTINATION third_party/glog)
   IF(ANDROID)
     INSTALL(FILES ${GLOG_LIBRARIES} DESTINATION third_party/glog/lib/${ANDROID_ABI})

cmake/external/protobuf.cmake

@@ -250,7 +250,7 @@ IF(NOT PROTOBUF_FOUND)
     SET(PROTOBUF_PROTOC_LIBRARY ${extern_protobuf_PROTOC_LIBRARY}
         CACHE FILEPATH "protoc library." FORCE)
 
-    IF(WITH_C_API OR WITH_FLUID)
+    IF(WITH_C_API)
         INSTALL(DIRECTORY ${PROTOBUF_INCLUDE_DIR} DESTINATION third_party/protobuf)
         IF(ANDROID)
             INSTALL(FILES ${PROTOBUF_LITE_LIBRARY} DESTINATION third_party/protobuf/lib/${ANDROID_ABI})

cmake/external/warpctc.cmake

@@ -52,6 +52,7 @@ ExternalProject_Add(
                     -DWITH_TORCH=OFF
                     -DCMAKE_DISABLE_FIND_PACKAGE_Torch=ON
                     -DBUILD_SHARED=ON
+                    -DBUILD_TESTS=OFF
                     -DCMAKE_POSITION_INDEPENDENT_CODE=ON
                     -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE}
                     ${EXTERNAL_OPTIONAL_ARGS}

cmake/generic.cmake

@@ -179,20 +179,24 @@ function(cc_library TARGET_NAME)
   set(oneValueArgs "")
   set(multiValueArgs SRCS DEPS)
   cmake_parse_arguments(cc_library "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
-  if (cc_library_SRCS)
-    if (cc_library_SHARED OR cc_library_shared) # build *.so
+  if(cc_library_SRCS)
+    if(cc_library_SHARED OR cc_library_shared) # build *.so
       add_library(${TARGET_NAME} SHARED ${cc_library_SRCS})
     else()
       add_library(${TARGET_NAME} STATIC ${cc_library_SRCS})
     endif()
-    if (cc_library_DEPS)
+    if(cc_library_DEPS)
       # Don't need link libwarpctc.so
-      if ("${cc_library_DEPS};" MATCHES "warpctc;")
+      if("${cc_library_DEPS};" MATCHES "warpctc;")
         list(REMOVE_ITEM cc_library_DEPS warpctc)
         add_dependencies(${TARGET_NAME} warpctc)
       endif()
+      # Support linking flags: --whole-archive (Linux) / -force_load (MacOS)
+      target_circle_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
+      if("${cc_library_DEPS}" MATCHES "ARCHIVE_START")
+        list(REMOVE_ITEM cc_library_DEPS ARCHIVE_START ARCHIVE_END)
+      endif()
       add_dependencies(${TARGET_NAME} ${cc_library_DEPS})
-      target_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
     endif()
     
     # cpplint code style

cmake/inference_lib.cmake

+# make package for paddle fluid shared and static library
+function(copy TARGET)
+    set(options "")
+    set(oneValueArgs "")
+    set(multiValueArgs SRCS DSTS DEPS)
+    cmake_parse_arguments(copy_lib "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+
+    list(LENGTH copy_lib_SRCS copy_lib_SRCS_len)
+    list(LENGTH copy_lib_DSTS copy_lib_DSTS_len)
+    if(NOT ${copy_lib_SRCS_len} EQUAL ${copy_lib_DSTS_len})
+        message(FATAL_ERROR "${TARGET} source numbers are not equal to destination numbers")
+    endif()
+    math(EXPR len "${copy_lib_SRCS_len} - 1")
+    
+    add_custom_target(${TARGET} DEPENDS ${copy_lib_DEPS})
+    foreach(index RANGE ${len})
+        list(GET copy_lib_SRCS ${index} src)
+        list(GET copy_lib_DSTS ${index} dst)
+        add_custom_command(TARGET ${TARGET} PRE_BUILD COMMAND mkdir -p "${dst}")
+        if(IS_DIRECTORY ${src})
+            add_custom_command(TARGET ${TARGET} PRE_BUILD COMMAND cp -r "${src}" "${dst}")
+        else()
+            add_custom_command(TARGET ${TARGET} PRE_BUILD COMMAND cp "${src}" "${dst}")
+        endif()
+    endforeach()
+endfunction()
+
+# third party
+set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/eigen3")
+copy(eigen3_lib
+  SRCS ${EIGEN_INCLUDE_DIR}/Eigen/Core ${EIGEN_INCLUDE_DIR}/Eigen/src ${EIGEN_INCLUDE_DIR}/unsupported/Eigen
+  DSTS ${dst_dir}/Eigen ${dst_dir}/Eigen ${dst_dir}/unsupported
+)
+
+set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/gflags")
+copy(gflags_lib
+  SRCS ${GFLAGS_INCLUDE_DIR} ${GFLAGS_LIBRARIES}
+  DSTS ${dst_dir} ${dst_dir}/lib
+)
+
+set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/glog")
+copy(glog_lib
+  SRCS ${GLOG_INCLUDE_DIR} ${GLOG_LIBRARIES}
+  DSTS ${dst_dir} ${dst_dir}/lib
+)
+
+IF(NOT PROTOBUF_FOUND)
+    set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/protobuf")
+    copy(protobuf_lib
+      SRCS ${PROTOBUF_INCLUDE_DIR} ${PROTOBUF_LITE_LIBRARY}
+      DSTS ${dst_dir} ${dst_dir}/lib
+    )
+ENDIF(NOT PROTOBUF_FOUND)
+
+# paddle fluid module
+set(src_dir "${PADDLE_SOURCE_DIR}/paddle")
+set(dst_dir "${CMAKE_INSTALL_PREFIX}/paddle")
+set(module "framework")
+copy(framework_lib DEPS framework_py_proto 
+  SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/details/*.h ${PADDLE_BINARY_DIR}/paddle/framework/framework.pb.h
+  DSTS ${dst_dir}/${module} ${dst_dir}/${module}/details ${dst_dir}/${module}
+)
+
+set(module "memory")
+copy(memory_lib
+  SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/detail/*.h
+  DSTS ${dst_dir}/${module} ${dst_dir}/${module}/detail
+)
+
+set(module "inference")
+copy(inference_lib DEPENDS paddle_fluid_shared
+  SRCS ${src_dir}/${module}/*.h ${PADDLE_BINARY_DIR}/paddle/inference/libpaddle_fluid.so
+  DSTS ${dst_dir}/${module} ${dst_dir}/${module}
+)
+
+set(module "platform")
+copy(platform_lib
+  SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/dynload/*.h ${src_dir}/${module}/details/*.h
+  DSTS ${dst_dir}/${module} ${dst_dir}/${module}/dynload ${dst_dir}/${module}/details
+)
+
+set(module "string")
+copy(string_lib
+  SRCS ${src_dir}/${module}/*.h ${src_dir}/${module}/tinyformat/*.h
+  DSTS ${dst_dir}/${module} ${dst_dir}/${module}/tinyformat
+)
+
+add_custom_target(inference_lib_dist DEPENDS 
+  inference_lib framework_lib memory_lib platform_lib string_lib
+  gflags_lib glog_lib protobuf_lib eigen3_lib)

doc/CMakeLists.txt

@@ -47,3 +47,5 @@ sphinx_add_target(paddle_docs_cn
                   ${SPHINX_CACHE_DIR_CN}
                   ${CMAKE_CURRENT_SOURCE_DIR}
                   ${SPHINX_HTML_DIR_CN})
+
+add_subdirectory(api)

doc/api/CMakeLists.txt

+# configured documentation tools and intermediate build results
+set(BINARY_BUILD_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/_build")
+
+# Sphinx cache with pickled ReST documents
+set(SPHINX_CACHE_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/_doctrees")
+
+# HTML output director
+set(SPHINX_HTML_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/html")
+
+configure_file(
+    "${CMAKE_CURRENT_SOURCE_DIR}/../templates/conf.py.en.in"
+    "${BINARY_BUILD_DIR_EN}/conf.py"
+    @ONLY)
+
+sphinx_add_target(paddle_api_docs
+                  html
+                  ${BINARY_BUILD_DIR_EN}
+                  ${SPHINX_CACHE_DIR_EN}
+                  ${CMAKE_CURRENT_SOURCE_DIR}
+                  ${SPHINX_HTML_DIR_EN})

doc/api/v2/fluid/layers.rst

@@ -323,6 +323,12 @@ batch_norm
 ..  autofunction:: paddle.v2.fluid.layers.batch_norm
     :noindex:
 
+layer_norm
+----------
+
+..  autofunction:: paddle.v2.fluid.layers.layer_norm
+    :noindex:
+
 beam_search_decode
 ------------------
 

doc/build_and_install/index_cn.rst

+安装与编译
+==========
+
+.. _install_steps:
+
+安装流程
+++++++++
+
+PaddlePaddle提供pip和Docker的安装方式：
+
+.. toctree::
+   :maxdepth: 1
+
+   pip_install_cn.rst
+   docker_install_cn.rst
+   build_cn.md
+
+编译流程
+++++++++
+
+..  warning::
+
+    建议直接使用上述安装流程，方便快速安装。只有在遇到需要独立定制的二进制时才需要编译。
+
+..  toctree::
+    :maxdepth: 1
+
+    build_from_source_cn.rst
+
+常见问题解答
+++++++++++
+
+`常见问题解答 <http://www.paddlepaddle.org/docs/develop/documentation/zh/faq/build_and_install/index_cn.html>`_

doc/build_and_install/index_en.rst

+Install and Build
+=================
+
+.. _install_steps:
+
+Install Steps
+++++++++
+
+You can choose either pip or Docker to complete your install:
+
+.. toctree::
+   :maxdepth: 1
+
+   pip_install_en.rst
+   docker_install_en.rst
+   build_en.md
+
+
+Build from Source
+-----------------
+
+..  warning::
+
+    We recommend to directly install via above installation steps, you'll only need to build PaddlePaddle from source when you need a modifed binary.
+
+..  toctree::
+    :maxdepth: 1
+
+    build_from_source_en.md
+
+FAQ
+++++++++++
+
+`FAQ <http://www.paddlepaddle.org/docs/develop/documentation/zh/faq/build_and_install/index_en.html>`_

doc/design/auto_gradient_check.md

-## Auto Gradient Checker Design
+## Auto Gradient Check Design
 
-## Backgraound：
-- Generally, it is easy to check whether the forward computation of an Operator is correct or not. However, backpropagation is a notoriously difficult algorithm to debug and get right:
-  1. you should get the right backpropagation formula according to the forward computation.
-  2. you should implement it right in CPP.
-  3. it's difficult to prepare test data.
+## Background：
+- Generally, it is easy to check whether the forward computation of an Operator is correct or not. However, backpropagation is a notoriously difficult algorithm to debug and get right because of the following challenges:
+  1. The formula for backpropagation formula should be correct according to the forward computation.
+  2. The Implementation of the above shoule be correct in CPP.
+  3. It is difficult to prepare an unbiased test data.
 
-- Auto gradient checking gets a numerical gradient by forward Operator and use it as a reference of the backward Operator's result. It has several advantages:
-  1. numerical gradient checker only need forward operator.
-  2. user only need to prepare the input data for forward Operator.
+- Auto gradient checking gets a numerical gradient using forward Operator and uses it as a reference for the backward Operator's result. It has several advantages:
+  1. Numerical gradient checker only needs the forward operator.
+  2. The user only needs to prepare the input data for forward Operator and not worry about the backward Operator.
 
 ## Mathematical Theory
-The following two document from Stanford has a detailed explanation of how to get numerical gradient and why it's useful.
+The following documents from Stanford have a detailed explanation of how to compute the numerical gradient and why it is useful.
 
 - [Gradient checking and advanced optimization(en)](http://deeplearning.stanford.edu/wiki/index.php/Gradient_checking_and_advanced_optimization)
 - [Gradient checking and advanced optimization(cn)](http://ufldl.stanford.edu/wiki/index.php/%E6%A2%AF%E5%BA%A6%E6%A3%80%E9%AA%8C%E4%B8%8E%E9%AB%98%E7%BA%A7%E4%BC%98%E5%8C%96)
 
 
-## Numeric Gradient Implementation
+## Numerical Gradient Implementation
 ### Python Interface
 ```python
 def get_numerical_gradient(op,
@@ -27,73 +27,76 @@ def get_numerical_gradient(op,
                          delta=0.005,
                          local_scope=None):
     """
-    Get Numeric Gradient for an operator's input.
+    Get Numerical Gradient for the input of an operator.
 
-    :param op: C++ operator instance, could be an network
+    :param op: C++ operator instance, could be an network.
     :param input_values: The input variables. Should be an dictionary, whose key is
-    variable name, and value is numpy array.
+    variable name, and value is a numpy array.
     :param output_name: The final output variable name.
-    :param input_to_check: The input variable with respect to which to compute the gradient.
-    :param delta: The perturbation value for numeric gradient method. The
-    smaller delta is, the more accurate result will get. But if that delta is
-     too small, it will suffer from numerical stability problem.
+    :param input_to_check: The input variable with respect to which the gradient has to be computed.
+    :param delta: The perturbation value for numerical gradient method. The
+    smaller the delta, the more accurate the result. But if the delta is too
+    small, it will suffer from the numerical stability problem.
     :param local_scope: The local scope used for get_numeric_gradient.
     :return: The gradient array in numpy format.
     """
 ```
 
-### Explaination:
+### Explanation:
 
-- Why need `output_name`
-  - An Operator may have multiple Output, one can get independent gradient from each Output. So caller should specify the name of the output variable.
+- Why do we need an `output_name`
+  - An Operator may have multiple Outputs, one can compute an independent gradient from each Output. So the caller should specify the name of the output variable.
 
-- Why need `input_to_check`
-  - One operator may have multiple inputs. Gradient Op can calculate the gradient of these inputs at the same time. But Numeric Gradient needs to calculate them one by one. So `get_numeric_gradient` is designed to calculate the gradient for one input. If you need to compute multiple inputs, you can call `get_numeric_gradient` multiple times.
+- Why do we need `input_to_check`
+  - One operator can have multiple inputs. Gradient Op can calculate the gradient of these inputs at the same time. But Numerical Gradient needs to calculate them one by one. So `get_numeric_gradient` is designed to calculate the gradient for one input. If you need to compute multiple inputs, you can call `get_numeric_gradient` multiple times each with a different input.
 
 
 ### Core Algorithm Implementation
 
 
 ```python
-    # we only compute gradient of one element a time.
+    # we only compute the gradient of one element a time.
     # we use a for loop to compute the gradient of each element.
     for i in xrange(tensor_size):
-        # get one input element by its index i.
-        origin = tensor_to_check.get_float_element(i)
+        # get one input element using the index i.
+        original = tensor_to_check.get_float_element(i)
 
-        # add delta to it, run op and then get the new value of the result tensor.
-        x_pos = origin + delta
+        # add delta to it, run the forward op and then
+        # get the new value of the result tensor.
+        x_pos = original + delta
         tensor_to_check.set_float_element(i, x_pos)
         y_pos = get_output()
 
-        # plus delta to this element, run op and get the new value of the result tensor.
-        x_neg = origin - delta
+        # Subtract delta from this element, run the op again
+        # and get the new value of the result tensor.
+        x_neg = original - delta
         tensor_to_check.set_float_element(i, x_neg)
         y_neg = get_output()
 
         # restore old value
-        tensor_to_check.set_float_element(i, origin)
+        tensor_to_check.set_float_element(i, original)
 
-        # compute the gradient of this element and store it into a numpy array.
+        # compute the gradient of this element and store
+        # it into a numpy array.
         gradient_flat[i] = (y_pos - y_neg) / delta / 2
 
     # reshape the gradient result to the shape of the source tensor.
     return gradient_flat.reshape(tensor_to_check.get_dims())
 ```
 
-## Auto Graident Checker Framework
+## Auto Gradient Check Framework
 
 Each Operator Kernel has three kinds of Gradient:
 
 1. Numerical gradient
 2. CPU kernel gradient
-3. GPU kernel gradient (if supported)
+3. GPU kernel gradient (if supported by the device)
 
-The numerical gradient only relies on forward Operator. So we use the numerical gradient as the reference value. And the gradient checking is performed in the following three steps:
+The numerical gradient only relies on the forward Operator, so we use the numerical gradient as the reference value. The gradient checking is performed in the following three steps:
 
-1. calculate the numerical gradient
-2. calculate CPU kernel gradient with the backward Operator and compare it with the numerical gradient
-3. calculate GPU kernel gradient with the backward Operator and compare it with the numeric gradient (if supported)
+1. Calculate the numerical gradient
+2. Calculate CPU kernel gradient with the backward Operator and compare it with the numerical gradient.
+3. Calculate GPU kernel gradient with the backward Operator and compare it with the numeric gradient. (if supported)
 
 #### Python Interface
 
@@ -110,25 +113,26 @@ The numerical gradient only relies on forward Operator. So we use the numerical
         :param forward_op: used to create backward_op
         :param input_vars: numpy value of input variable. The following
           computation will use these variables.
-        :param inputs_to_check: the input variable with respect to which to compute the gradient.
+        :param inputs_to_check: the input variable with respect to which the
+          gradient will be computed.
         :param output_name: The final output variable name.
         :param max_relative_error: The relative tolerance parameter.
-        :param no_grad_set: used when create backward ops
+        :param no_grad_set: used to create backward ops
         :param only_cpu: only compute and check gradient on cpu kernel.
         :return:
         """
 ```
 
-### How to check if two numpy array is close enough?
-if `abs_numerical_grad` is nearly zero, then use abs error for numerical_grad
+### How to check if two numpy arrays are close enough?
+if `abs_numerical_grad` is nearly zero, then use absolute error for numerical_grad.
 
 ```python
 numerical_grad = ...
 operator_grad = numpy.array(scope.find_var(grad_var_name(name)).get_tensor())
 
 abs_numerical_grad = numpy.abs(numerical_grad)
-# if abs_numerical_grad is nearly zero, then use abs error for numeric_grad, not relative
-# error.
+# if abs_numerical_grad is nearly zero, then use abs error for
+# numeric_grad, instead of relative error.
 abs_numerical_grad[abs_numerical_grad < 1e-3] = 1
 
 diff_mat = numpy.abs(abs_numerical_grad - operator_grad) / abs_numerical_grad
@@ -140,7 +144,7 @@ max_diff = numpy.max(diff_mat)
 The Input data for auto gradient checker should be reasonable to avoid numerical stability problem.
 
 
-#### Refs:
+#### References:
 
 - [Gradient checking and advanced optimization(en)](http://deeplearning.stanford.edu/wiki/index.php/Gradient_checking_and_advanced_optimization)
 - [Gradient checking and advanced optimization(cn)](http://ufldl.stanford.edu/wiki/index.php/%E6%A2%AF%E5%BA%A6%E6%A3%80%E9%AA%8C%E4%B8%8E%E9%AB%98%E7%BA%A7%E4%BC%98%E5%8C%96)

doc/design/cpp_data_feeding.md

+# C++ Data Feeding
+
+In training with Paddle V2 API, data feeding wholly dependents on Python code. To get rid of the Python environment and achieve the goal of "wrapping the whole training by a while loop op" in Paddle Fluid, a C++ data feeding mechanism is required. 
+
+In this document we show the fundamental design of C++ data feeding process, which includes the data reading, shuffling and batching.
+
+## Reader
+
+A new concept named 'Reader' is introduced. `Reader` is a series of inherited classes which can be hold by our `Variable` and they are used to read or process file data.
+
+
+### `ReaderBase`
+
+`ReaderBase` is the abstract base class of all readers. It defines the all readers' interfaces.
+
+```cpp
+class ReaderBase {
+ public:
+  explicit ReaderBase(const std::vector<DDim>& shapes) : shapes_(shapes) {
+    PADDLE_ENFORCE(!shapes_.empty());
+  }
+  // Read the next batch of data. (A 'batch' can be only one instance)
+  virtual void ReadNext(std::vector<LoDTensor>* out) = 0;
+  // Show whether the next bacth exists.
+  virtual bool HasNext() const = 0;
+  
+  // Reinitialize the reader and read the file from the begin.
+  virtual void ReInit() = 0;
+  
+  // Get a certain read in data's shape.
+  DDim shape(size_t idx) const;
+  // Get shapes of all read in data.
+  std::vector<DDim> shapes() const { return shapes_; }
+  // Set shapes of read in data.
+  void set_shapes(const std::vector<DDim>& shapes) { shapes_ = shapes; }
+
+  virtual ~ReaderBase() {}
+
+ protected:
+  std::vector<DDim> shapes_;
+};
+```
+
+### `FileReader` and `DecoratedReader`
+
+These two classes are derived from the `ReaderBase` and will further be derived by respective specific readers. That is to say, in our design, there are two kinds of readers: file readers and decorated readers. A file reader reads from a file of some specific format, and yield only one instance of data at a time. e.g. RecordIO reader, jpg reader, .... A decorated reader takes another reader(both file reader and decorated reader are OK) as its 'underlying reader'. It gets data from its underlying reader, does some process on them(shuffling, or batching), then yields processed data. The output data of a decorated reader can be a single instance or a batch. `ShuffleReader` and `BatchReader` are both decorated readers.
+
+All the readers share exactly the same interfaces defined in `ReaderBase`. So they can be decorated for more than one time: We can **shuffle** a reader's outputs and then **batch** the shuffle outputs. The interface consistency also allows related ops use readers without knowing what they are exactly.
+
+
+### `ReaderHolder`
+
+Different readers belong to different class types. It leads to a problem: How can we drop them into `Variable`s and fetch them out by a unified method? For example, if a Variable holds a `BatchReader`, we can not get it by the following code:
+
+```cpp
+var->Get<ReaderBase>("batch_reader");
+```
+
+we have to write:
+
+```cpp
+var->Get<BatchReader>("batch_reader");
+```
+
+This requires each time getting a reader from a variable we must know the reader's type exactly. It is nearly impossible.
+
+To solve this problem, we introduce `ReaderHolder` as a wrapper. It acts as an empty decorator of `ReaderBase`, which erases reader's type. With `ReaderHolder` we are able to fetch all types of readers by `var->Get<ReaderHolder>("...")` and regard the obtained object as a reader.
+
+## Related Operators
+
+To create and invoke readers, some now ops are introduced:
+
+### `CreateReaderOp`
+
+Each reader has its creating op. File readers' creating ops have no input and yield the created file reader as its output. Decorated readers' creating ops take the underlying readers as inputs and then yield new decorated readers.
+
+### `ReadOp`
+
+A reader is only a Variable. It cannot trigger the reading process by itself. So we add the `ReadOp` to execute it. A `ReadOp` takes a reader Variable as its input. Each time it runs, it invokes the reader‘s `ReadNext()` function and gets a new batch of data(or only one instance of data, if we use file reader directly). The output data of a reader are in the form of `std::vector<LoDTenosr>`, so the `ReadOp` also needs to split the vector and move LoDTensors to their respective output Variables.

doc/design/csp.md

@@ -42,7 +42,7 @@ The type *channel* is conceptually the blocking queue.  In Go, its implemented i
 
 The `select` operation has been in OS kernels long before Go language.  All Unix kernels implement system calls *poll* and *select*.  They monitor multiple file descriptors to see if I/O is possible on any of them.  This takes O(N) time.  Since Linux 2.6, a new system call, *epoll*, can do the same in O(1) time.  In BSD systems, there is a similar system call *kqueue*.  Go's Linux implementation uses epoll.
 
-It might be a good idea to implement Fluid's select using epoll too.  In this design doc, we start from the O(N) way, so we could focus on Python binding and the syntax.
+It might be a good idea to implement Fluid's select using epoll too.  In this design doc, we start from the O(N) way so that we could focus on Python binding and the syntax.
 
 ### Type Channel
 
@@ -71,14 +71,14 @@ ch1 := make(chan int, 100)  // a channel that can buffer 100 ints.
 In Fluid, we should be able to do the same:
 
 ```python
-ch  = fluid.make_chan(dtype=INT)
-ch1 = fluid.make_chan(dtype=INT, 100)
+ch  = fluid.make_channel(dtype=INT)
+ch1 = fluid.make_channel(dtype=INT, 100)
 ```
 
 In addition to that, we want channels that can hold more complex element types, e.g., Tensors of float16:
 
 ```python
-ch = fluid.make_chan(dtype=Tensor, etype=float16)
+ch = fluid.make_channel(dtype=Tensor, etype=float16)
 ```
 
 or Tensors of Tensors of float16 etc.
@@ -87,8 +87,136 @@ The point here is that we need a consistent way to compose types, like in C++ we
 
 ### Send and Recv
 
+Go's CSP implementation depends on data type *channel*. There are two types of channels:
+
+1. The unblocked channel, or buffered channel, is a blocking queue with a non-zero sized buffer. The sending to buffered channel blocks if the buffer is full, and the receive operation blocks if the buffer is empty.
+1. blocked channel, or unbuffered channel, is a blocking queue with no buffer.  Both sending and receiving block with unbuffered channels.
+
+There are four types of actions with a channel:
+
+1. Create a channel
+
+   ```go
+   ch := make(chan int) // this is an unbuffered channel
+   ch := make(chan int, 100) // this is a buffered channel of 100 ints.
+   ```
+
+1. Send
+
+   ```go
+   ch <- 111
+   ```
+
+1. Recv
+
+   ```go
+   y, ok <- ch
+   ```
+
+1. Close
+
+   ```go
+   close(ch)
+   ```
+   
+   Please be aware that a closed channel is not a nil channel, which is `var ch chan int`.
+   
+There are some [axioms with channels](https://dave.cheney.net/2014/03/19/channel-axioms):
+
+1. A send to a nil channel blocks forever
+
+1. A receive from a nil channel blocks forever
+
+1. A send to a closed channel panics
+
+1. A receive from a closed channel returns the residual values and then zeros.
+
+In Fluid, we have [buffered channels](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/details/buffered_channel.h) and [unbuffered channels](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/details/unbuffered_channel.h)
+
+The following program illustrates the Python syntax for accessing Fluid buffers.
+
+```python
+import fluid
+
+buffer_size = 10
+ch = fluid.make_channel(dtype=INT, buffer_size)
+
+# Now write three elements to the channel
+with fluid.while(steps=buffer_size):
+  fluid.send(ch, step)
+
+fluid.close_channel(ch)
+
+with fluid.while(steps=buffer_size):
+  fluid.print(fluid.recv(ch))
+```
+
+The following example shows that to avoid the always-blocking behavior of unbuffered channels, we need to use Fluid's goroutines.
+
+```python
+import fluid
+
+ch = fluid.make_channel(dtype=INT)
+
+with fluid.go():
+  fluid.send(ch)
+
+y = fluid.recv(ch)
+
+fluid.close_channel(ch)
+```
+
 ### Select
 
+In Go, the `select` statement lets a goroutine wait on multiple communication operations. A `select` blocks until one of its cases can run, then it executes that case. It chooses one at random if multiple are ready.
+
+```go
+
+ch1  := make(chan int)       
+ch2  := make(chan int, 100)
+
+x := 0
+
+for {
+    select {
+    case ch1 <- x:
+      x := x + 1
+    case y <- ch2:
+      fmt.Println("Received on channel")
+    default:
+      fmt.Println("Default")
+    }
+  }
+
+```
+
+In Fluid, we should be able to do the same:
+
+```python
+ch1  = fluid.make_chan(dtype=INT)
+ch2 = fluid.make_chan(dtype=INT, 100)
+
+sel = fluid.select()
+
+with sel.case(ch1, 'w', X):
+    fluid.layers.increment(X)
+
+with sel.case(ch2, 'r', Y):
+    fluid.print("Received on Channel")
+
+with sel.default():
+    fluid.print("Default")
+
+```
+
+In the above code snippet, `X` and `Y` are variables. Now let us look at each of these statements one by one.
+
+- `sel.case(ch1, 'w', X)` : This specifies that we are writing to `ch1` and we want to write the integer in variable `X` to the channel. The character `w` is used here to make the syntax familiar to write syntax in Python I/O.
+
+- `sel.case(ch2, 'r', Y)` : This specifies that we would like to read the result from `ch2` into variable `Y`. The character `r` is used here to make the syntax familiar to read syntax in Python I/O.
+
+- `sel.default()` : This is equivalent to the default in Go `select`. If none of the channels are ready for read or write, then the fluid code in the default block will be executed.
+
 ## Example Programs
 
 ### 1. RPC between Trainers and Parameter Servers

doc/design/switch.md

@@ -10,8 +10,7 @@ The following example shows the usage of `fluid.switch`.
 a = fluid.Var(10)
 b = fluid.Var(0)
 
-switch = fluid.switch()
-with switch.block():
+with switch() as switch:
     with switch.case(fluid.less_equal(a, 10)):
         fluid.print("Case 1")
     with switch.case(fluid.larger(a, 0)):

doc/dev/contribute_to_paddle_en.md

+../../CONTRIBUTING.md
\ No newline at end of file

doc/dev/index_cn.rst

+开发标准
+========
+
+..  toctree::
+  :maxdepth: 1
+
+  contribute_to_paddle_cn.md
+  write_docs_cn.rst

doc/dev/index_en.rst

+Development
+------------
+
+..  toctree::
+  :maxdepth: 1
+
+  new_layer_en.rst
+  contribute_to_paddle_en.md
+  write_docs_en.rst

doc/dev/write_docs_cn.rst

+#############
+如何贡献文档
+#############
+
+PaddlePaddle的文档包括英文文档 ``doc`` 和中文文档 ``doc_cn`` 两个部分。文档都是通过 `cmake`_ 驱动 `sphinx`_ 编译生成，生成后的文档分别存储在编译目录的 ``doc`` 和 ``doc_cn`` 两个子目录下。
+也可以利用PaddlePaddle 工具来编译文档，这个情况下所有的文件会存在整理过的的文件目录 .ppo_workspace/content 下
+
+如何构建文档
+============
+
+PaddlePaddle的文档构建有三种方式。
+
+
+使用PaddlePaddle.org工具
+--------------
+这个是目前推荐的使用方法。除了可以自动编译文档，也可以直接在网页预览文档。
+
+文件工具是使用Docker，需要在系统里先安装好Docker工具包。Docker安装请参考Docker的官网。安装好Docker之后及可用以下命令启动工具
+
+..  code-block:: bash
+
+    mkdir paddlepaddle # Create paddlepaddle working directory
+    cd paddlepaddle
+
+    # Clone the content repositories
+    git clone https://github.com/PaddlePaddle/Paddle.git
+    git clone https://github.com/PaddlePaddle/book.git
+    git clone https://github.com/PaddlePaddle/models.git
+    git clone https://github.com/PaddlePaddle/Mobile.git
+
+    # Please specify the working directory through -v
+    docker run -it -p 8000:8000 -v `pwd`:/var/content paddlepaddle/paddlepaddle.org:latest
+
+注意: PaddlePaddle.org 会在 -v (volume) 指定的内容存储库运行命令
+之后再用网页连到http://localhost:8000就可以在网页上生成需要的文档
+编译后的文件将被存储在工作目录 <paddlepaddle working directory>/.ppo_workspace/content。
+
+如果不想使用 Docker，你还可以通过运行Django框架直接激活工具的服务器。使用下面的命令来运行它。
+
+..  code-block:: bash
+
+    mkdir paddlepaddle # Create paddlepaddle working directory
+    cd paddlepaddle
+
+    # Clone the content repositories and PaddlePaddle.org
+    git clone https://github.com/PaddlePaddle/Paddle.git
+    git clone https://github.com/PaddlePaddle/book.git
+    git clone https://github.com/PaddlePaddle/models.git
+    git clone https://github.com/PaddlePaddle/Mobile.git
+    git clone https://github.com/PaddlePaddle/PaddlePaddle.org.git
+
+    # Please specify the PaddlePaddle working directory. In the current setting, it should be pwd
+    export CONTENT_DIR=<path_to_paddlepaddle_working_directory>
+    export ENV=''
+    cd PaddlePaddle.org/portal/
+    pip install -r requirements.txt
+    python manage.py runserver
+
+工具服务器将读取环境变量 CONTENT_DIR 搜索代码库。请指定的PaddlePaddle工作目录给环境变量 CONTENT_DIR。
+之后再用网页连到http://localhost:8000就可以在网页上生成需要的文档。
+编译后的文件将被存储在工作目录 <paddlepaddle working directory>/.ppo_workspace/content。
+
+想了解更多PaddlePaddle.org工具的详细信息，可以 `点击这里 <https://github.com/PaddlePaddle/PaddlePaddle.org/blob/develop/README.cn.md>`_ 。
+
+使用Docker构建
+--------------
+
+使用Docker构建PaddlePaddle的文档，需要在系统里先安装好Docker工具包。Docker安装请参考 `Docker的官网 <https://docs.docker.com/>`_ 。安装好Docker之后可以使用源码目录下的脚本构建文档，即
+
+..  code-block:: bash
+
+    cd TO_YOUR_PADDLE_CLONE_PATH
+    cd paddle/scripts/tools/build_docs
+    sh build_docs.sh
+
+编译完成之后，会在当前目录生成两个子目录\: doc(英文文档目录)和 doc_cn(中文文档目录)。
+打开浏览器访问对应目录下的index.html即可访问本地文档。
+
+直接构建
+--------
+
+如果提示正确，可以执行以下命令编译生成文档，即
+
+..  code-block:: bash
+
+    cd TO_YOUR_PADDLE_CLONE_PATH
+    mkdir -p build
+    cd build
+    cmake .. -DCMAKE_BUILD_TYPE=Debug -DWITH_GPU=OFF -DWITH_MKL=OFF -DWITH_DOC=ON
+    make gen_proto_py
+    make paddle_docs paddle_docs_cn
+
+编译完成之后，会在当前目录生成两个子目录\: doc(英文文档目录)和 doc_cn(中文文档目录)。
+打开浏览器访问对应目录下的index.html即可访问本地文档。
+
+
+如何书写文档
+============
+
+PaddlePaddle文档使用 `sphinx`_ 自动生成，用户可以参考sphinx教程进行书写。
+
+如何更新www.paddlepaddle.org
+============================
+
+更新的文档以PR的形式提交到github中，提交方式参见 `贡献文档 <http://www.paddlepaddle.org/docs/develop/documentation/en/howto/dev/contribute_to_paddle_en.html>`_ 。
+目前PaddlePaddle的develop分支的文档是自动触发更新的，用户可以分别查看最新的 `中文文档 <http://www.paddlepaddle.org/docs/develop/documentation/zh/getstarted/index_cn.html>`_ 和
+`英文文档 <http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/index_en.html>`_ 。
+
+
+..  _cmake: https://cmake.org/
+..  _sphinx: http://www.sphinx-doc.org/en/1.4.8/

doc/dev/write_docs_en.rst

+########################
+Contribute Documentation
+########################
+
+PaddlePaddle supports English documentation ``doc`` and Chinese documentation ``doc_cn``.
+Both are compiled by `cmake`_ and `sphinx`_ , the compiled documentations will be stored under ``doc`` and ``doc_cn`` directories.
+When using the PaddlePaddle.org to compile documentations, the compiled documentations will be stored under a consolidated directory: .ppo_workspace/content
+
+How to Build Documentations
+============
+
+We recommend using PaddlePaddle.org tool to build documentation
+
+
+Use PaddlePaddle.org tool
+--------------
+This is the recommended method to build documentation. It can compile documentation and preview the documentation in a web browser.
+
+The tool uses Docker, please install it on your system. Please check Docker official website on how to install Docker. You may use the following commands to activate the tool
+
+..  code-block:: bash
+
+    mkdir paddlepaddle # Create paddlepaddle working directory
+    cd paddlepaddle
+
+    # Clone the content repositories. You may only clone the contents you need
+    git clone https://github.com/PaddlePaddle/Paddle.git
+    git clone https://github.com/PaddlePaddle/book.git
+    git clone https://github.com/PaddlePaddle/models.git
+    git clone https://github.com/PaddlePaddle/Mobile.git
+
+    # Please specify the working directory through -v
+    docker run -it -p 8000:8000 -v `pwd`:/var/content paddlepaddle/paddlepaddle.org:latest
+
+Note: PaddlePaddle.org will read the content repos specified in the -v (volume) flag of the docker run command
+Use a web browser and navigate to http://localhost:8000, click the buttons to compile the documentation
+The compiled documentations will be stored in <paddlepaddle working directory>/.ppo_workspace/content
+
+
+If you don't wish to use Docker, you can also activate the tool through Django. Use the following the commands to set up
+
+..  code-block:: bash
+
+    mkdir paddlepaddle # Create paddlepaddle working directory
+    cd paddlepaddle
+
+    # Clone the content repositories and PaddlePaddle.org
+    git clone https://github.com/PaddlePaddle/Paddle.git
+    git clone https://github.com/PaddlePaddle/book.git
+    git clone https://github.com/PaddlePaddle/models.git
+    git clone https://github.com/PaddlePaddle/Mobile.git
+    git clone https://github.com/PaddlePaddle/PaddlePaddle.org.git
+
+    # Please specify the PaddlePaddle working directory. In the current setting, it should be pwd
+    export CONTENT_DIR=<path_to_paddlepaddle_working_directory>
+    export ENV=''
+    cd PaddlePaddle.org/portal/
+    pip install -r requirements.txt
+    python manage.py runserver
+
+Use a web browser and navigate to http://localhost:8000, click the buttons to compile the documentation
+The compiled documentations will be stored in <paddlepaddle working directory>/.ppo_workspace/content
+
+If you want to learn more on the PaddlePaddle.org, please `click here <https://github.com/PaddlePaddle/PaddlePaddle.org/blob/develop/README.md>`_ 。
+
+How to write Documentations
+============
+
+PaddlePaddle uses `sphinx`_ to compile documentations，Please check sphinx official website for more detail.
+
+
+How to update www.paddlepaddle.org
+============================
+
+Please create PRs and submit them to github, please check `Contribute Code <http://www.paddlepaddle.org/docs/develop/documentation/en/howto/dev/contribute_to_paddle_en.html>`_ 。
+PaddlePaddle develop branch will update the documentation once the PR is merged. User may check latest `Chinese Docs <http://www.paddlepaddle.org/docs/develop/documentation/zh/getstarted/index_cn.html>`_ and
+`English Docs <http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/index_en.html>`_ 。
+
+..  _cmake: https://cmake.org/
+..  _sphinx: http://www.sphinx-doc.org/en/1.4.8/

doc/getstarted/build_and_install/index_cn.rst

-安装与编译
-==========
-
-.. _install_steps:
-
-安装流程
-++++++++
-
-PaddlePaddle提供pip和Docker的安装方式：
-
-.. toctree::
-   :maxdepth: 1
-
-   pip_install_cn.rst
-   docker_install_cn.rst
-   ../../howto/dev/build_cn.md
-
-编译流程
-++++++++
-
-..  warning::
-
-    建议直接使用上述安装流程，方便快速安装。只有在遇到需要独立定制的二进制时才需要编译。
-
-..  toctree::
-    :maxdepth: 1
-
-    build_from_source_cn.rst
-
-常见问题解答
-++++++++++
-
-`常见问题解答 <http://www.paddlepaddle.org/docs/develop/documentation/zh/faq/build_and_install/index_cn.html>`_

doc/getstarted/build_and_install/index_en.rst

-Install and Build
-=================
-
-.. _install_steps:
-
-Install Steps
-++++++++
-
-You can choose either pip or Docker to complete your install:
-
-.. toctree::
-   :maxdepth: 1
-
-   pip_install_en.rst
-   docker_install_en.rst
-   ../../howto/dev/build_en.md
-
-
-Build from Source
------------------
-
-..  warning::
-
-    We recommend to directly install via above installation steps, you'll only need to build PaddlePaddle from source when you need a modifed binary.
-
-..  toctree::
-    :maxdepth: 1
-
-    build_from_source_en.md
-
-FAQ
-++++++++++
-
-`FAQ <http://www.paddlepaddle.org/docs/develop/documentation/zh/faq/build_and_install/index_en.html>`_

doc/getstarted/concepts/use_concepts_cn.rst

@@ -4,7 +4,7 @@
 
 PaddlePaddle是源于百度的一个深度学习平台。PaddlePaddle为深度学习研究人员提供了丰富的API，可以轻松地完成神经网络配置，模型训练等任务。
 这里将介绍PaddlePaddle的基本使用概念，并且展示了如何利用PaddlePaddle来解决一个经典的线性回归问题。
-在使用该文档之前，请参考 `安装文档 <../build_and_install/index_cn.html>`_ 完成PaddlePaddle的安装。
+在使用该文档之前，请参考 `安装文档 <../../build_and_install/index_cn.html>`_ 完成PaddlePaddle的安装。
 
 
 配置网络

doc/getstarted/index_cn.rst

 新手入门
 ============
 
-.. _quick_install:
-
-快速安装
-++++++++
-
-PaddlePaddle支持使用pip快速安装，目前支持CentOS 6以上, Ubuntu 14.04以及MacOS 10.12，并安装有Python2.7。
-执行下面的命令完成快速安装，版本为cpu_avx_openblas：
-
-  .. code-block:: bash
-
-     pip install paddlepaddle
-
-如果需要安装支持GPU的版本（cuda7.5_cudnn5_avx_openblas），需要执行：
-
-  .. code-block:: bash
-
-     pip install paddlepaddle-gpu
-
-更详细的安装和编译方法参考：
-
-..  toctree::
-  :maxdepth: 1
-
-  build_and_install/index_cn.rst
-
-.. _quick_start:
-
-快速开始
-++++++++
-
-创建一个 housing.py 并粘贴此Python代码：
-
-  .. code-block:: python
-
-     import paddle.v2 as paddle
-
-     # Initialize PaddlePaddle.
-     paddle.init(use_gpu=False, trainer_count=1)
-
-     # Configure the neural network.
-     x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(13))
-     y_predict = paddle.layer.fc(input=x, size=1, act=paddle.activation.Linear())
-
-     # Infer using provided test data.
-     probs = paddle.infer(
-         output_layer=y_predict,
-         parameters=paddle.dataset.uci_housing.model(),
-         input=[item for item in paddle.dataset.uci_housing.test()()])
-
-     for i in xrange(len(probs)):
-         print 'Predicted price: ${:,.2f}'.format(probs[i][0] * 1000)
-
-执行 :code:`python housing.py` 瞧！ 它应该打印出预测住房数据的清单。
-
 ..  toctree::
   :maxdepth: 1
 
+  quickstart_cn.rst
   concepts/use_concepts_cn.rst

doc/getstarted/index_en.rst

 GET STARTED
 ============
 
-.. _quick_install:
-
-Quick Install
-----------------------
-
-You can use pip to install PaddlePaddle with a single command, supports
-CentOS 6 above, Ubuntu 14.04 above or MacOS 10.12, with Python 2.7 installed.
-Simply run the following command to install, the version is cpu_avx_openblas:
-
-  .. code-block:: bash
-
-     pip install paddlepaddle
-
-If you need to install GPU version (cuda7.5_cudnn5_avx_openblas), run:
-
-  .. code-block:: bash
-
-     pip install paddlepaddle-gpu
-
-For more details about installation and build:
-
 ..  toctree::
   :maxdepth: 1
 
-  build_and_install/index_en.rst
-
-
-.. _quick_start:
-
-Quick Start
-++++++++
-
-Create a new file called housing.py, and paste this Python
-code:
-
-
-  .. code-block:: python
-
-     import paddle.v2 as paddle
-
-     # Initialize PaddlePaddle.
-     paddle.init(use_gpu=False, trainer_count=1)
-
-     # Configure the neural network.
-     x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(13))
-     y_predict = paddle.layer.fc(input=x, size=1, act=paddle.activation.Linear())
-
-     # Infer using provided test data.
-     probs = paddle.infer(
-         output_layer=y_predict,
-         parameters=paddle.dataset.uci_housing.model(),
-         input=[item for item in paddle.dataset.uci_housing.test()()])
-
-     for i in xrange(len(probs)):
-         print 'Predicted price: ${:,.2f}'.format(probs[i][0] * 1000)
-
-Run :code:`python housing.py` and voila! It should print out a list of predictions
-for the test housing data.
+  quickstart_en.rst

doc/getstarted/quickstart_cn.rst

+快速开始
+========
+
+快速安装
+--------
+
+PaddlePaddle支持使用pip快速安装，目前支持CentOS 6以上, Ubuntu 14.04以及MacOS 10.12，并安装有Python2.7。
+执行下面的命令完成快速安装，版本为cpu_avx_openblas：
+
+  .. code-block:: bash
+
+     pip install paddlepaddle
+
+如果需要安装支持GPU的版本（cuda7.5_cudnn5_avx_openblas），需要执行：
+
+  .. code-block:: bash
+
+     pip install paddlepaddle-gpu
+
+更详细的安装和编译方法参考：:ref:`install_steps` 。
+
+快速使用
+--------
+
+创建一个 housing.py 并粘贴此Python代码：
+
+  .. code-block:: python
+
+     import paddle.v2 as paddle
+
+     # Initialize PaddlePaddle.
+     paddle.init(use_gpu=False, trainer_count=1)
+
+     # Configure the neural network.
+     x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(13))
+     y_predict = paddle.layer.fc(input=x, size=1, act=paddle.activation.Linear())
+
+     # Infer using provided test data.
+     probs = paddle.infer(
+         output_layer=y_predict,
+         parameters=paddle.dataset.uci_housing.model(),
+         input=[item for item in paddle.dataset.uci_housing.test()()])
+
+     for i in xrange(len(probs)):
+         print 'Predicted price: ${:,.2f}'.format(probs[i][0] * 1000)
+
+执行 :code:`python housing.py` 瞧！ 它应该打印出预测住房数据的清单。

doc/getstarted/quickstart_en.rst

+Quick Start
+============
+
+Quick Install
+-------------
+
+You can use pip to install PaddlePaddle with a single command, supports
+CentOS 6 above, Ubuntu 14.04 above or MacOS 10.12, with Python 2.7 installed.
+Simply run the following command to install, the version is cpu_avx_openblas:
+
+  .. code-block:: bash
+
+     pip install paddlepaddle
+
+If you need to install GPU version (cuda7.5_cudnn5_avx_openblas), run:
+
+  .. code-block:: bash
+
+     pip install paddlepaddle-gpu
+
+For more details about installation and build: :ref:`install_steps` .
+
+Quick Use
+---------
+
+Create a new file called housing.py, and paste this Python
+code:
+
+
+  .. code-block:: python
+
+     import paddle.v2 as paddle
+
+     # Initialize PaddlePaddle.
+     paddle.init(use_gpu=False, trainer_count=1)
+
+     # Configure the neural network.
+     x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(13))
+     y_predict = paddle.layer.fc(input=x, size=1, act=paddle.activation.Linear())
+
+     # Infer using provided test data.
+     probs = paddle.infer(
+         output_layer=y_predict,
+         parameters=paddle.dataset.uci_housing.model(),
+         input=[item for item in paddle.dataset.uci_housing.test()()])
+
+     for i in xrange(len(probs)):
+         print 'Predicted price: ${:,.2f}'.format(probs[i][0] * 1000)
+
+Run :code:`python housing.py` and voila! It should print out a list of predictions
+for the test housing data.

doc/howto/capi/compile_paddle_lib_cn.md

+## 安装与编译C-API预测库
+
+### 概述
+
+使用 C-API 进行预测依赖于将 PaddlePaddle 核心代码编译成链接库，只需在编译时需配制下面这些编译选项：
+
+必须配置选项：
+- `WITH_C_API`，必须配置为`ON`。
+
+推荐配置选项：
+- `WITH_PYTHON`，推荐配置为`OFF`
+- `WITH_SWIG_PY`，推荐配置为`OFF`
+- `WITH_GOLANG`，推荐设置为`OFF`
+
+可选配置选项：
+- `WITH_GPU`，可配置为`ON/OFF`
+- `WITH_MKL`，可配置为`ON/OFF`
+
+对推荐配置中的选项建议按照设置，以避免链接不必要的库。其它可选编译选项按需进行设定。
+
+下面的代码片段从github拉取最新代码，配制编译选项（需要将PADDLE_ROOT替换为PaddlePaddle预测库的安装路径）：
+
+```shell
+PADDLE_ROOT=/path/of/capi
+git clone https://github.com/PaddlePaddle/Paddle.git
+cd Paddle
+mkdir build
+cd build
+cmake -DCMAKE_INSTALL_PREFIX=$PADDLE_ROOT \
+      -DCMAKE_BUILD_TYPE=Release \
+      -DWITH_C_API=ON \
+      -DWITH_SWIG_PY=OFF \
+      -DWITH_GOLANG=OFF \
+      -DWITH_PYTHON=OFF \
+      -DWITH_MKL=OFF \
+      -DWITH_GPU=OFF  \
+      ..
+```
+
+执行上述代码生成Makefile文件后，执行：`make && make install`。成功编译后，使用C-API所需的依赖（包括：（1）编译出的PaddlePaddle预测库和头文件；（2）第三方链接库和头文件）均会存放于`PADDLE_ROOT`目录中。
+
+编译成功后在 `PADDLE_ROOT` 下会看到如下目录结构（包括了编译出的PaddlePaddle头文件和链接库，以及第三方依赖链接库和头文件（如果需要，由链接方式决定））：
+
+```text
+├── include
+│   └── paddle
+│       ├── arguments.h
+│       ├── capi.h
+│       ├── capi_private.h
+│       ├── config.h
+│       ├── error.h
+│       ├── gradient_machine.h
+│       ├── main.h
+│       ├── matrix.h
+│       ├── paddle_capi.map
+│       └── vector.h
+├── lib
+│   ├── libpaddle_capi_engine.a
+│   ├── libpaddle_capi_layers.a
+│   ├── libpaddle_capi_shared.so
+│   └── libpaddle_capi_whole.a
+└── third_party
+    ├── gflags
+    │   ├── include
+    │   │   └── gflags
+    │   │       ├── gflags_completions.h
+    │   │       ├── gflags_declare.h
+    │   │       ...
+    │   └── lib
+    │       └── libgflags.a
+    ├── glog
+    │   ├── include
+    │   │   └── glog
+    │   │       ├── config.h
+    │   │       ...
+    │   └── lib
+    │       └── libglog.a
+    ├── openblas
+    │   ├── include
+    │   │   ├── cblas.h
+    │   │   ...
+    │   └── lib
+    │       ...
+    ├── protobuf
+    │   ├── include
+    │   │   └── google
+    │   │       └── protobuf
+    │   │           ...
+    │   └── lib
+    │       └── libprotobuf-lite.a
+    └── zlib
+        ├── include
+        │   ...
+        └── lib
+            ...
+
+```
+
+### 链接说明
+
+目前提供三种链接方式：
+
+1. 链接`libpaddle_capi_shared.so` 动态库
+    - 使用 PaddlePaddle C-API 开发预测程序链接`libpaddle_capi_shared.so`时，需注意：
+        1. 如果编译时指定编译CPU版本，且使用`OpenBLAS`数学库，在使用C-API开发预测程序时，只需要链接`libpaddle_capi_shared.so`这一个库。
+        1. 如果是用编译时指定CPU版本，且使用`MKL`数学库，由于`MKL`库有自己独立的动态库文件，在使用PaddlePaddle C-API开发预测程序时，需要自己链接MKL链接库。
+        1. 如果编译时指定编译GPU版本，CUDA相关库会在预测程序运行时动态装载，需要将CUDA相关的库设置到`LD_LIBRARY_PATH`环境变量中。
+    - 这种方式最为简便，链接相对容易，**在无特殊需求情况下，推荐使用此方式**。
+
+2. 链接静态库 `libpaddle_capi_whole.a`
+    - 使用PaddlePaddle C-API 开发预测程序链接`libpaddle_capi_whole.a`时，需注意：
+        1. 需要指定`-Wl,--whole-archive`链接选项。
+        1. 需要显式地链接 `gflags`、`glog`、`libz`、`protobuf` 等第三方库，可在`PADDLE_ROOT/third_party`下找到。
+        1. 如果在编译 C-API 时使用OpenBLAS数学库，需要显示地链接`libopenblas.a`。
+        1. 如果在编译 C-API 是使用MKL数学库，需要显示地链接MKL的动态库。
+
+3. 链接静态库 `libpaddle_capi_layers.a`和`libpaddle_capi_engine.a`
+    - 使用PaddlePaddle C-API 开发预测程序链接`libpaddle_capi_whole.a`时，需注意：
+        1. 这种链接方式主要用于移动端预测。
+        1. 为了减少生成链接库的大小把`libpaddle_capi_whole.a`拆成以上两个静态链接库。
+        1. 需指定`-Wl,--whole-archive -lpaddle_capi_layers` 和 `-Wl,--no-whole-archive -lpaddle_capi_engine` 进行链接。
+        1. 第三方依赖库需要按照与方式2同样方法显示地进行链接。

doc/howto/capi/index_cn.rst

+C-API预测库
+==================
+
+..  toctree::
+  :maxdepth: 1
+
+  compile_paddle_lib_cn.md
+  organization_of_the_inputs_cn.md
+  workflow_of_capi_cn.md

doc/howto/capi/workflow_of_capi_cn.md

+## C-API使用流程
+
+这篇文档介绍 PaddlePaddle C-API 整体使用流程。
+
+### 使用流程
+
+使用 C-API 的工作流程如图1所示，分为（1）准备预测模型和（2）预测程序开发两大部分。
+
+<p align="center">
+<img src="https://user-images.githubusercontent.com/5842774/34658453-365f73ea-f46a-11e7-9b3f-0fd112b27bae.png" width=500><br> 图1. C-API使用流程示意图
+</p>
+
+- 准备预测模型
+    1. 只将神经网络结构进行序列化。
+        - 只对神经网络结构进行序列化，加载模型需同时指定：网络结构的序列化结果和模型参数存储目录。
+    1. 将网络结构定义和训练结束存储下来的模型参数文件（多个）合并入一个文件。
+        - 神经网络模型结构和训练好的模型将被序列化合并入一个文件。
+        - 预测时只需加载一个文件便于发布。
+    - **注意**：以上两种方式只需选择其一即可。
+- 调用 C-API 开发预测序
+    1. 初始化PaddlePaddle运行环境。
+    1. 加载预测模型。
+    1. 创建神经网络输入，组织输入数据。
+    1. 进行前向计算，获得计算结果。
+    1. 清理和结束。
+
+### 准备预测模型
+
+准备预测模型部分，我们以手写数字识别任务为例进行介绍。手写数字识别任务定义了一个含有[两个隐层的简单全连接网络](https://github.com/PaddlePaddle/book/blob/develop/02.recognize_digits/README.cn.md#softmax回归softmax-regression)，网络接受一幅图片作为输入，将图片分类到 0 ~ 9 类别标签之一。完整代码可以查看[此目录](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense) 中的相关脚本。
+
+调用C-API开发预测程序需要一个训练好的模型，运行[MNIST手写数字识别目录](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense)下的[mnist_v2.py](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/examples/model_inference/dense/mnist_v2.py)脚本，在终端执行`python mnist_v2.py`，会使用 PaddlePaddle 内置的 [MNIST 数据集](http://yann.lecun.com/exdb/mnist/)进行训练。训练好的模型默认保存在当前运行目录下的`models`目录中。
+
+下面，我们将训练结束后存储下来的模型转换成预测模型。
+
+1. 序列化神经网络模型配置
+
+    PaddlePaddle 使用 protobuf 来传输网络配置文件中定义的网络结构和相关参数，使用 C-API 进行预测时，需要将网络结构使用 protobuf 进行序列化，写入文件中。
+
+    调用[`paddle.utils.dump_v2_config`](https://github.com/PaddlePaddle/Paddle/tree/develop/python/paddle/utils/dump_v2_config.py)中的`dump_v2_config`函数能够将使用 PaddlePaddle V2 API 定义的神经网络结构 dump 到指定文件中，示例代码如下：
+
+    ```python
+    from paddle.utils.dump_v2_config import dump_v2_config
+    from mnist_v2 import network
+
+    predict = network(is_infer=True)
+    dump_v2_config(predict, "trainer_config.bin", True)
+    ```
+
+    对[手写数字识别](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense)这个示例，[`mnist_v2.py`](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense/mnist_v2.py)脚本集成了序列化神经网络结构的过程，可以直接运行 `python mnist_v2.py --task dump_config` 对神经网络结构进行序列化，结果会写入当前运行目录下的`trainer_config.bin`文件中。
+
+    使用这种方式，需要**在运行时将神经网络的多个可学习参数放在同一个目录中**，C-API可以通过分别指定序列化后的网络结构文件和参数目录来加载训练好的模型。
+
+2. 合并模型文件(可选)
+
+    一些情况为了便于发布，希望能够将序列化后的神经网络结构和训练好的模型参数打包进一个文件。对于这样的需求，可以使用`paddle.utils.merge_model`中的`merge_v2_model`接口对神经网络结构和训练好的参数进行序列化，将序列化结果写入一个文件内。
+
+    代码示例如下：
+
+    ```python
+    from paddle.utils.merge_model import merge_v2_modelss
+    from mnist_v2 import network
+
+    net = network(is_infer=True)
+    param_file = "models/params_pass_4.tar"
+    output_file = "output.paddle.model"
+    merge_v2_model(net, param_file, output_file)
+    ```
+    对[手写数字识别](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense)这个示例，可直接运行 `python` [merge_v2_model.py](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference/dense/merge_v2_model.py)。序列化结果会写入当前运行目录下的`output.paddle.model`文件中。使用这种方式，运行时C-API可以通过指定`output.paddle.model`文件的路径来加载预测模型。
+
+#### 注意事项
+1. 为使用C-API，在调用`dump_v2_config`序列化神经网络结构时，参数`binary`必须指定为`True`。
+1. **预测使用的网络结构往往不同于训练**，通常需要去掉网络中的：（1）类别标签层；（2）损失函数层；（3）`evaluator`等，只留下核心计算层，请注意是否需要修改网络结构。
+1. 预测时，可以获取网络中定义的任意多个（大于等于一个）层前向计算的结果，需要哪些层的计算结果作为输出，就将这些层加入一个Python list中，作为调用`dump_v2_config`的第一个参数。
+
+### 编写预测代码
+
+预测代码更多详细示例代码请参考[C-API使用示例](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/capi/examples/model_inference) 目录下的代码示例。这一节对图1中预测代码编写的5个步骤进行介绍和说明。
+
+#### step 1. 初始化PaddlePaddle运行环境
+第一步需调用[`paddle_init`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/main.h#L27) 初始化PaddlePaddle运行环境，该接口接受两个参数：参数的个数和参数列表。
+
+#### step2. 加载模型
+
+这里介绍C-API使用中的一个重要概念：Gradient Machine。
+
+概念上，在 PaddlePaddle 内部，一个GradientMachine类的对象管理着一组计算层（PaddlePaddle Layers）来完成前向和反向计算，并处理与之相关的所有细节。在调用C-API预测时，只需进行前向计算而无需调用反向计算。这篇文档之后部分会使用`gradient machine`来特指调用PaddlePaddle C-API创建的GradientMachine类的对象。每一个 `gradient machine` 都会管理维护一份训练好的模型，下面是C-API提供的，两种常用的模型加载方式：
+
+1. 调用[`paddle_gradient_machine_load_parameter_from_disk`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/gradient_machine.h#L61)接口，从磁盘加载预测模型。这时`gradient machine`会独立拥有一份训练好的模型；
+1. 调用[`paddle_gradient_machine_create_shared_param`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/gradient_machine.h#L88)接口，与其它`gradient machine`的共享已经加载的预测模型。这种情况多出现在使用多线程预测时，通过多个线程共享同一个模型来减少内存开销。可参考[此示例](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/examples/model_inference/multi_thread/main.c)。
+
+- 注意事项
+    1. 使用PaddlePaddle V2 API训练，模型中所有可学习参数会被存为一个压缩文件，需要手动进行解压，将它们放在同一目录中，C-API不会直接加载 V2 API 存储的压缩文件。
+    1. 如果使用`merge model`方式将神经网络结构和训练好的参数序列化到一个文件，请参考此[示例](https://github.com/PaddlePaddle/Mobile/blob/develop/Demo/linux/paddle_image_recognizer.cpp#L59)。
+    1. 通过灵活使用以上两个接口，加载模型可其它多种方式，例如也可在程序运行过程中再加载另外一个模型。
+
+#### step 3. 创建神经网络输入，组织输入数据
+
+基本使用概念：
+- 在PaddlePaddle内部，神经网络中一个计算层的输入输出被组织为一个 `Argument` 结构体，如果神经网络有多个输入或者多个输出，每一个输入/输出都会对应有自己的`Argument`。
+- `Argument` 并不真正“存储”数据，而是将输入/输出数据有机地组织在一起。
+- 在`Argument`内部由：1. `Matrix`（二维矩阵，存储浮点类型输入/输出）；2. `IVector`（一维数组，**仅用于存储整型值**，多用于自然语言处理任务）来实际存储数据。
+
+C-API支持的所有输入数据类型和他们的组织方式，请参考“输入/输出数据组织”一节。
+
+这篇文档的之后部分会使用`argument`来特指PaddlePaddle C-API中神经网络的一个输入/输出，使用`paddle_matrix`**特指**`argument`中用于存储数据的`Matrix`类的对象。
+
+在组织神经网络输入，获取输出时，需要思考完成以下工作：
+1. 为每一个输入/输出创建`argument`；
+1. 为每一个`argument`创建`paddle_matrix`来存储数据；
+
+与输入不同的是，不需在使用C-API时为输出`argument`的`paddle_matrix`对象分配空间。前向计算之后PaddlePaddle内部已经分配/管理了每个计算层输出的存储空间。
+
+#### step 4. 前向计算
+
+完成上述准备之后，通过调用 [`paddle_gradient_machine_forward`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/capi/gradient_machine.h#L73) 接口完成神经网络的前向计算。
+
+#### step 5. 清理
+
+结束预测之后，对使用的中间变量和资源进行清理和释放。

doc/howto/cluster/cmd_argument_cn.md

+## 启动参数说明
+
+下面以`doc/howto/cluster/src/word2vec`中的代码作为实例，介绍使用PaddlePaddle v2 API完成分布式训练。
+
+### 启动参数服务器
+执行以下的命令启动一个参数服务器并等待和计算节点的数据交互
+```bash
+$ paddle pserver --port=7164 --ports_num=1 --ports_num_for_sparse=1 --num_gradient_servers=1
+```
+
+如果希望可以在后台运行pserver程序，并保存输出到一个日志文件，可以运行：
+```bash
+$ stdbuf -oL /usr/bin/nohup paddle pserver --port=7164 --ports_num=1 --ports_num_for_sparse=1 --num_gradient_servers=1 &> pserver.log
+```
+
+参数说明
+
+- port：**必选，默认7164**，pserver监听的起始端口，根据ports_num决定总端口个数，从起始端口监听多个端口用于通信
+- ports_num：**必选，默认1**，监听的端口个数
+- ports_num_for_sparse：**必选，默认0**，用于稀疏类型参数通信的端口个数
+- num_gradient_servers：**必选，默认1**，当前训练任务pserver总数
+
+### 启动计算节点
+执行以下命令启动使用python编写的trainer程序（文件名为任意文件名，如train.py）
+```bash
+$ python train.py
+```
+
+trainer需要和pserver保持网络联通以完成训练。trainer启动需要传入端口、pserver地址等参数使trainer可以正确连接到pserver。这些参数可以通过[环境变量](https://zh.wikipedia.org/wiki/环境变量)或编写程序时`paddle.init()`中传入参数。如果同时使用`paddle.init()`参数和环境变量，将会优先使用`paddle.init()`中传入的参数。
+
+使用环境变量：
+
+```bash
+export PADDLE_INIT_USE_GPU=False
+export PADDLE_INIT_TRAINER_COUNT=1
+export PADDLE_INIT_PORT=7164
+export PADDLE_INIT_PORTS_NUM=1
+export PADDLE_INIT_PORTS_NUM_FOR_SPARSE=1
+export PADDLE_INIT_NUM_GRADIENT_SERVERS=1
+export PADDLE_INIT_TRAINER_ID=0
+export PADDLE_INIT_PSERVERS=127.0.0.1
+```
+
+使用参数：
+
+```python
+paddle.init(
+        use_gpu=False,
+        trainer_count=1,
+        port=7164,
+        ports_num=1,
+        ports_num_for_sparse=1,
+        num_gradient_servers=1,
+        trainer_id=0,
+        pservers="127.0.0.1")
+```
+
+参数说明
+
+- use_gpu： **可选，默认False**，是否启用GPU训练
+- trainer_count：**必选，默认1**，当前trainer的线程数目
+- port：**必选，默认7164**，连接到pserver的端口
+- ports_num：**必选，默认1**，连接到pserver的端口个数
+- ports_num_for_sparse：**必选，默认0**，和pserver之间用于稀疏类型参数通信的端口个数
+- num_gradient_servers：**必选，默认1**，当前训练任务trainer总数
+- trainer_id：**必选，默认0**，每个trainer的唯一ID，从0开始的整数
+- pservers：**必选，默认127.0.0.1**，当前训练任务启动的pserver的IP列表，多个IP使用“,”隔开
+
+
+### 准备数据集
+
+参考样例数据准备脚本[prepare.py](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/howto/usage/cluster/src/word2vec/prepare.py)，准备训练数据和验证数据集，我们使用paddle.dataset.imikolov数据集，并根据分布式训练并发数（trainer节点个数），在`prepare.py`开头部分指定`SPLIT_COUNT`将数据切分成多份。
+
+在线上系统中，通常会使用MapReduce任务的输出结果作为训练结果，这样训练文件的个数会比较多，而且个数并不确定。在trainer中可以使用下面取模的方法为每个trainer分配训练数据文件：
+
+```python
+import os
+train_list = []
+flist = os.listdir("/train_data/")
+for f in flist:
+  suffix = int(f.split("-")[1])
+  if suffix % TRAINER_COUNT == TRAINER_ID:
+    train_list.append(f)
+```
+
+示例程序`prepare.py`会把训练集和测试集分别分割成多个文件（例子中为3个，后缀为`-00000`、`-00001`和`-00002`）:
+```
+train.txt
+train.txt-00000
+train.txt-00001
+train.txt-00002
+test.txt
+test.txt-00000
+test.txt-00001
+test.txt-00002
+```
+
+在进行分布式训练时，每个trainer进程需要能够读取属于自己的一份数据。在一些分布式系统中，系统会提供一个分布式存储服务，这样保存在分布式存储中的数据可以被集群中的每个节点读取到。如果不使用分布式存储，则需要手动拷贝属于每个trainer节点的训练数据到对应的节点上。
+
+对于不同的训练任务，训练数据格式和训练程序的`reader()`会大不相同，所以开发者需要根据自己训练任务的实际场景完成训练数据的分割和`reader()`的编写。
+
+### 准备训练程序
+
+我们会对每个训练任务都会在每个节点上创建一个工作空间（workspace），其中包含了用户的训练程序、程序依赖、挂载或下载的训练数据分片。
+
+最后，工作空间应如下所示：
+```
+.
+|-- my_lib.py
+|-- word_dict.pickle
+|-- train.py
+|-- train_data_dir/
+|   |-- train.txt-00000
+|   |-- train.txt-00001
+|   |-- train.txt-00002
+`-- test_data_dir/
+    |-- test.txt-00000
+    |-- test.txt-00001
+    `-- test.txt-00002
+```
+
+- `my_lib.py`：会被`train.py`调用的一些用户定义的库函数，比如PIL库等。
+- `word_dict.pickle`：在`train.py`中会使用到的字典数据文件。
+- `train.py`：训练程序，代码参考[api_train_v2_cluster.py](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/howto/usage/cluster/src/word2vec/api_train_v2_cluster.py)。***注意：*** 对于本样例代码，在使用不同的分布式计算平台时，您可能需要修改`train.py`开头的部分（如下），以便获得训练数据的位置和获取环境变量配置：
+
+  ```python
+  cluster_train_file = "./train_data_dir/train/train.txt"
+  cluster_test_file = "./test_data_dir/test/test.txt"
+  node_id = os.getenv("OMPI_COMM_WORLD_RANK")
+  if not node_id:
+      raise EnvironmentError("must provied OMPI_COMM_WORLD_RANK")
+  ```
+
+- `train_data_dir`：包含训练数据的目录，可以是从分布式存储挂载过来的，也可以是在任务启动前下载到本地的。
+- `test_data_dir`：包含测试数据集的目录。

doc/howto/cluster/cmd_argument_en.md

+## Command-line arguments
+
+We'll take `doc/howto/cluster/src/word2vec` as an example to introduce distributed training using PaddlePaddle v2 API.
+
+### Starting parameter server
+
+Type the below command to start a parameter server which will wait for trainers to connect:
+
+```bash
+$ paddle pserver --port=7164 --ports_num=1 --ports_num_for_sparse=1 --num_gradient_servers=1
+```
+
+If you wish to run parameter servers in background, and save a log file, you can type:
+```bash
+$ stdbuf -oL /usr/bin/nohup paddle pserver --port=7164 --ports_num=1 --ports_num_for_sparse=1 --num_gradient_servers=1 &> pserver.log
+```
+
+Parameter Description
+
+- port: **required, default 7164**, port which parameter server will listen on. If ports_num greater than 1, parameter server will listen on multiple ports for more network throughput.
+- ports_num: **required, default 1**, total number of ports will listen on.
+- ports_num_for_sparse: **required, default 0**, number of ports which serves sparse parameter update.
+- num_gradient_servers: **required, default 1**, total number of gradient servers.
+
+### Starting trainer
+Type the command below to start the trainer(name the file whatever you want, like "train.py")
+
+```bash
+$ python train.py
+```
+
+Trainers' network need to be connected with parameter servers' network to finish the job. Trainers need to know port and IPs to locate parameter servers. You can pass arguments to trainers through [environment variables](https://en.wikipedia.org/wiki/Environment_variable) or pass to `paddle.init()` function. Arguments passed to the `paddle.init()` function will overwrite environment variables.
+
+Use environment viriables:
+
+```bash
+export PADDLE_INIT_USE_GPU=False
+export PADDLE_INIT_TRAINER_COUNT=1
+export PADDLE_INIT_PORT=7164
+export PADDLE_INIT_PORTS_NUM=1
+export PADDLE_INIT_PORTS_NUM_FOR_SPARSE=1
+export PADDLE_INIT_NUM_GRADIENT_SERVERS=1
+export PADDLE_INIT_TRAINER_ID=0
+export PADDLE_INIT_PSERVERS=127.0.0.1
+python train.py
+```
+
+Pass arguments:
+
+```python
+paddle.init(
+        use_gpu=False,
+        trainer_count=1,
+        port=7164,
+        ports_num=1,
+        ports_num_for_sparse=1,
+        num_gradient_servers=1,
+        trainer_id=0,
+        pservers="127.0.0.1")
+```
+
+Parameter Description
+
+- use_gpu: **optional, default False**, set to "True" to enable GPU training.
+- trainer_count: **required, default 1**, number of threads in current trainer.
+- port: **required, default 7164**, port to connect to parameter server.
+- ports_num: **required, default 1**, number of ports for communication.
+- ports_num_for_sparse: **required, default 0**, number of ports for sparse type caculation.
+- num_gradient_servers: **required, default 1**, number of trainers in current job.
+- trainer_id: **required, default 0**, ID for every trainer, start from 0.
+- pservers: **required, default 127.0.0.1**, list of IPs of parameter servers, separated by ",".
+
+### Prepare Training Dataset
+
+Here's some example code [prepare.py](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/howto/usage/cluster/src/word2vec/prepare.py), it will download public `imikolov` dataset and split it into multiple files according to job parallelism(trainers count). Modify `SPLIT_COUNT` at the begining of `prepare.py` to change the count of output files.
+
+In the real world, we often use `MapReduce` job's output as training data, so there will be lots of files. You can use `mod` to assign training file to trainers:
+
+```python
+import os
+train_list = []
+flist = os.listdir("/train_data/")
+for f in flist:
+  suffix = int(f.split("-")[1])
+  if suffix % TRAINER_COUNT == TRAINER_ID:
+    train_list.append(f)
+```
+
+Example code `prepare.py` will split training data and testing data into 3 files with digital suffix like `-00000`, `-00001` and`-00002`:
+
+```
+train.txt
+train.txt-00000
+train.txt-00001
+train.txt-00002
+test.txt
+test.txt-00000
+test.txt-00001
+test.txt-00002
+```
+
+When job started, every trainer needs to get it's own part of data. In some distributed systems a storage service will be provided, so the date under that path can be accessed by all the trainer nodes. Without the storage service, you must copy the training data to each trainer node.
+
+Different training jobs may have different data format and `reader()` function, developers may need to write different data prepare scripts and `reader()` functions for their job.
+
+### Prepare Training program
+
+We'll create a *workspace* directory on each node, storing your training program, dependencies, mounted or downloaded dataset directory.
+
+
+Your workspace may looks like:
+```
+.
+|-- my_lib.py
+|-- word_dict.pickle
+|-- train.py
+|-- train_data_dir/
+|   |-- train.txt-00000
+|   |-- train.txt-00001
+|   |-- train.txt-00002
+`-- test_data_dir/
+    |-- test.txt-00000
+    |-- test.txt-00001
+    `-- test.txt-00002
+```
+
+- `my_lib.py`: user defined libraries, like PIL libs. This is optional.
+- `word_dict.pickle`: dict file for training word embeding.
+- `train.py`: training program. Sample code: [api_train_v2_cluster.py](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/howto/usage/cluster/src/word2vec/api_train_v2_cluster.py). ***NOTE:*** You may need to modify the head part of `train.py` when using different cluster platform to retrive configuration environment variables:
+
+  ```python
+  cluster_train_file = "./train_data_dir/train/train.txt"
+  cluster_test_file = "./test_data_dir/test/test.txt"
+  node_id = os.getenv("OMPI_COMM_WORLD_RANK")
+  if not node_id:
+      raise EnvironmentError("must provied OMPI_COMM_WORLD_RANK")
+  ```
+
+- `train_data_dir`: containing training data. Mount from storage service or copy trainning data to here.
+- `test_data_dir`: containing testing data.

doc/howto/cluster/index_cn.rst

+分布式训练
+==========
+
+本节将介绍如何使用PaddlePaddle在不同的集群框架下完成分布式训练。分布式训练架构如下图所示：
+
+.. image:: src/ps_cn.png
+   :width: 500
+
+- 数据分片（Data shard): 用于训练神经网络的数据，被切分成多个部分，每个部分分别给每个trainer使用。
+- 计算节点（Trainer）: 每个trainer启动后读取切分好的一部分数据，开始神经网络的“前馈”和“后馈”计算，并和参数服务器通信。在完成一定量数据的训练后，上传计算得出的梯度（gradients），然后下载优化更新后的神经网络参数（parameters）。
+- 参数服务器（Parameter server）:每个参数服务器只保存整个神经网络所有参数的一部分。参数服务器接收从计算节点上传的梯度，并完成参数优化更新，再将更新后的参数下发到每个计算节点。
+
+这样，通过计算节点和参数服务器的分布式协作，可以完成神经网络的SGD方法的训练。PaddlePaddle可以同时支持同步随机梯度下降（SGD）和异步随机梯度下降。
+
+在使用同步SGD训练神经网络时，PaddlePaddle使用同步屏障（barrier），使梯度的提交和参数的更新按照顺序方式执行。在异步SGD中，则并不会等待所有trainer提交梯度才更新参数，这样极大地提高了计算的并行性：参数服务器之间不相互依赖，并行地接收梯度和更新参数，参数服务器也不会等待计算节点全部都提交梯度之后才开始下一步，计算节点之间也不会相互依赖，并行地执行模型的训练。可以看出，虽然异步SGD方式会提高参数更新并行度, 但是并不能保证参数同步更新，在任意时间某一台参数服务器上保存的参数可能比另一台要更新，与同步SGD相比，梯度会有噪声。
+
+..  toctree::
+  :maxdepth: 1
+
+  preparations_cn.md
+  cmd_argument_cn.md
+  multi_cluster/index_cn.rst

doc/howto/cluster/index_en.rst

+Distributed Training
+====================
+
+In this section, we'll explain how to run distributed training jobs with PaddlePaddle on different types of clusters. The diagram below shows the main architecture of a distributed trainning job:
+
+.. image:: src/ps_en.png
+   :width: 500
+
+- Data shard: training data will be split into multiple partitions, trainers use the partitions of the whole dataset to do the training job.
+- Trainer: each trainer reads the data shard, and train the neural network. Then the trainer will upload calculated "gradients" to parameter servers, and wait for parameters to be optimized on the parameter server side. When that finishes, the trainer download optimized parameters and continues its training.
+- Parameter server: every parameter server stores part of the whole neural network model data. They will do optimization calculations when gradients are uploaded from trainers, and then send updated parameters to trainers.
+
+PaddlePaddle can support both synchronize stochastic gradient descent (SGD) and asynchronous SGD.
+
+When training with synchronize SGD, PaddlePaddle uses an internal "synchronize barrier" which makes gradients update and parameter download in strict order. On the other hand, asynchronous SGD won't wait for all trainers to finish upload at a single step, this will increase the parallelism of distributed training: parameter servers do not depend on each other, they'll do parameter optimization concurrently. Parameter servers will not wait for trainers, so trainers will also do their work concurrently. But asynchronous SGD will introduce more randomness and noises in the gradient.
+
+..  toctree::
+  :maxdepth: 1
+
+  preparations_en.md
+  cmd_argument_en.md
+  multi_cluster/index_en.rst

doc/howto/cluster/multi_cluster/index_cn.rst

+在不同集群中运行
+================
+
+PaddlePaddle可以使用多种分布式计算平台构建分布式计算任务，包括：
+- `Kubernetes <http://kubernetes.io>`_ Google开源的容器集群的调度框架，支持大规模集群生产环境的完整集群方案。
+- `OpenMPI <https://www.open-mpi.org>`_ 成熟的高性能并行计算框架。
+- `Fabric <http://www.fabfile.org>`_ 集群管理工具。可以使用`Fabric`编写集群任务提交和管理脚本。
+
+对于不同的集群平台，会分别介绍集群作业的启动和停止方法。这些例子都可以在 `cluster_train_v2 <https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/scripts/cluster_train_v2>`_ 找到。
+
+在使用分布式计算平台进行训练时，任务被调度在集群中时，分布式计算平台通常会通过API或者环境变量提供任务运行需要的参数，比如节点的ID、IP和任务节点个数等。
+
+..  toctree::
+  :maxdepth: 1
+
+  fabric_cn.md
+  openmpi_cn.md
+  k8s_cn.md
+  k8s_distributed_cn.md
+  k8s_aws_cn.md

doc/howto/cluster/multi_cluster/index_en.rst

+Use different clusters
+======================
+
+PaddlePaddle supports running jobs on several platforms including:
+- `Kubernetes <http://kubernetes.io>`_ open-source system for automating deployment, scaling, and management of containerized applications from Google.
+- `OpenMPI <https://www.open-mpi.org>`_ Mature high performance parallel computing framework.
+- `Fabric <http://www.fabfile.org>`_ A cluster management tool. Write scripts to submit jobs or manage the cluster.
+
+We'll introduce cluster job management on these platforms. The examples can be found under `cluster_train_v2 <https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/scripts/cluster_train_v2>`_ .
+
+These cluster platforms provide API or environment variables for training processes, when the job is dispatched to different nodes. Like node ID, IP or total number of nodes etc.
+
+..  toctree::
+  :maxdepth: 1
+
+  fabric_en.md
+  openmpi_en.md
+  k8s_en.md
+  k8s_aws_en.md

doc/howto/cluster/preparations_cn.md

+## 环境准备
+
+1. 准备您的计算集群。计算集群通常由一组（几台到几千台规模）的Linux服务器组成。服务器之间可以通过局域网（LAN）联通，每台服务器具有集群中唯一的IP地址（或者可被DNS解析的主机名）。集群中的每台计算机通常被成为一个“节点”。
+1. 我们需要在集群的所有节点上安装 PaddlePaddle。 如果要启用GPU，还需要在节点上安装对应的GPU驱动以及CUDA。PaddlePaddle的安装可以参考[build_and_install](http://www.paddlepaddle.org/docs/develop/documentation/zh/getstarted/build_and_install/index_cn.html)的多种安装方式。我们推荐使用[Docker](http://www.paddlepaddle.org/docs/develop/documentation/zh/getstarted/build_and_install/docker_install_cn.html)安装方式来快速安装PaddlePaddle。
+
+安装完成之后，执行下面的命令可以查看已经安装的版本（docker安装方式可以进入docker容器执行：`docker run -it paddlepaddle/paddle:[tag] /bin/bash`）：
+```bash
+$ paddle version
+PaddlePaddle 0.10.0, compiled with
+    with_avx: ON
+    with_gpu: OFF
+    with_double: OFF
+    with_python: ON
+    with_rdma: OFF
+    with_timer: OFF
+```

doc/howto/cluster/preparations_en.md

+## Preparations
+
+1. Prepare your computer cluster. It's normally a bunch of Linux servers connected by LAN. Each server will be assigned a unique IP address. The computers in the cluster can be called "nodes".
+2. Install PaddlePaddle on every node. If you are going to take advantage of GPU cards, you'll also need to install proper driver and CUDA libraries. To install PaddlePaddle please read [this build and install](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/index_en.html) document. We strongly recommend using [Docker installation](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/docker_install_en.html).
+
+After installation, you can check the version by typing the below command (run a docker container  if using docker: `docker run -it paddlepaddle/paddle:[tag] /bin/bash`):
+
+```bash
+$ paddle version
+PaddlePaddle 0.10.0rc, compiled with
+    with_avx: ON
+    with_gpu: OFF
+    with_double: OFF
+    with_python: ON
+    with_rdma: OFF
+    with_timer: OFF
+```