Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into fix_warpctc_op

.copyright.hook

@@ -9,7 +9,7 @@ import subprocess
 import platform
 
 COPYRIGHT = '''
-  Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
+Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.

CMakeLists.txt

@@ -25,7 +25,6 @@ message(STATUS "CXX compiler: ${CMAKE_CXX_COMPILER}, version: "
 message(STATUS "C compiler: ${CMAKE_C_COMPILER}, version: "
         "${CMAKE_C_COMPILER_ID} ${CMAKE_C_COMPILER_VERSION}")
 
-find_package(Sphinx)
 if(NOT CMAKE_CROSSCOMPILING)
     find_package(CUDA QUIET)
 endif(NOT CMAKE_CROSSCOMPILING)
@@ -226,5 +225,7 @@ if(WITH_PYTHON)
 endif()
 
 if(WITH_DOC)
+    find_package(Sphinx REQUIRED)
+    find_python_module(recommonmark REQUIRED)
     add_subdirectory(doc)
 endif()

Dockerfile

@@ -70,7 +70,7 @@ RUN localedef -i en_US -f UTF-8 en_US.UTF-8
 # specify sphinx version as 1.5.6 and remove -U option for [pip install -U
 # sphinx-rtd-theme] since -U option will cause sphinx being updated to newest
 # version(1.7.1 for now), which causes building documentation failed.
-RUN pip install --upgrade pip==9.0.3 && \
+RUN easy_install -U pip && \
     pip install -U wheel && \
     pip install -U docopt PyYAML sphinx==1.5.6 && \
     pip install sphinx-rtd-theme==0.1.9 recommonmark

README.md

@@ -62,9 +62,9 @@ Please refer to our [release announcement](https://github.com/PaddlePaddle/Paddl
 ## Installation
 
 It is recommended to check out the
-[Docker installation guide](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/docker_install_en.html)
+[Docker installation guide](http://www.paddlepaddle.org/docs/develop/documentation/fluid/en/build_and_install/docker_install_en.html)
 before looking into the
-[build from source guide](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/build_from_source_en.html).
+[build from source guide](http://www.paddlepaddle.org/docs/develop/documentation/fluid/en/build_and_install/build_from_source_en.html).
 
 ## Documentation
 

benchmark/cluster/vgg16/vgg16_fluid.py

@@ -80,6 +80,8 @@ parser.add_argument(
     type=str,
     default="",
     help="Comma-separated list of hostname:port pairs")
+parser.add_argument(
+    "--profile", action='store_true', help="If set, profile a few steps.")
 
 # Flags for defining the tf.train.Server
 parser.add_argument(
@@ -183,8 +185,8 @@ def main():
             start_time = time.time()
             num_samples = 0
             train_pass_acc.reset()
-            for batch_id, data in enumerate(train_reader()):
-                ts = time.time()
+
+            def run_step(batch_id, data):
                 img_data = np.array(
                     map(lambda x: x[0].reshape(data_shape), data)).astype(
                         "float32")
@@ -196,14 +198,28 @@ def main():
                     feed={"pixel": img_data,
                           "label": y_data},
                     fetch_list=[avg_cost, batch_acc, batch_size])
+                return loss, acc, b_size
+
+            if args.profile and args.task_index == 0:
+                # warmup.
+                for batch_id, data in enumerate(train_reader()):
+                    if batch_id > 5: break
+                    run_step(batch_id, data)
+                with profiler.profiler('All', 'total', '/tmp/profile_vgg'):
+                    for batch_id, data in enumerate(train_reader()):
+                        if batch_id > 5: break
+                        run_step(batch_id, data)
+
+            for batch_id, data in enumerate(train_reader()):
+                ts = time.time()
+                loss, acc, b_size = run_step(batch_id, data)
                 iters += 1
                 num_samples += len(data)
                 train_pass_acc.add(value=acc, weight=b_size)
                 print(
-                    "Task:%d Pass = %d, Iters = %d, Loss = %f, Accuracy = %f, "
-                    "Speed = %.2f img/s " % (args.task_index, pass_id, iters,
-                                             loss, acc,
-                                             len(data) / (time.time() - ts))
+                    "Pass = %d, Iters = %d, Loss = %f, Accuracy = %f, "
+                    "Speed = %.2f img/s" % (pass_id, iters, loss, acc,
+                                            len(data) / (time.time() - ts))
                 )  # The accuracy is the accumulation of batches, but not the current batch.
 
             pass_elapsed = time.time() - start_time

benchmark/fluid/mnist.py

@@ -159,6 +159,7 @@ def run_benchmark(model, args):
         paddle.dataset.mnist.train(), batch_size=args.batch_size)
 
     accuracy = fluid.metrics.Accuracy()
+    train_exe = fluid.ParallelExecutor(use_cuda=True, loss_name=avg_cost.name)
     iters, num_samples, start_time = 0, 0, time.time()
     for pass_id in range(args.pass_num):
         accuracy.reset()
@@ -175,17 +176,20 @@ def run_benchmark(model, args):
             y_data = np.array(map(lambda x: x[1], data)).astype("int64")
             y_data = y_data.reshape([len(y_data), 1])
 
-            outs = exe.run(
-                fluid.default_main_program(),
+            outs = train_exe.run(
                 feed={"pixel": img_data,
                       "label": y_data},
-                fetch_list=[avg_cost, batch_acc, batch_size_tensor]
+                fetch_list=[
+                    avg_cost.name, batch_acc.name, batch_size_tensor.name
+                ]
             )  # The accuracy is the accumulation of batches, but not the current batch.
-            accuracy.update(value=outs[1], weight=outs[2])
+            accuracy.update(
+                value=np.array(np.mean(outs[1])),
+                weight=np.mean(np.array(outs[2])))
             iters += 1
             num_samples += len(y_data)
-            loss = np.array(outs[0])
-            acc = np.array(outs[1])
+            loss = np.mean(np.array(outs[0]))
+            acc = np.mean(np.array(outs[1]))
             train_losses.append(loss)
             train_accs.append(acc)
             print("Pass: %d, Iter: %d, Loss: %f, Accuracy: %f" %

benchmark/fluid/resnet.py

@@ -241,6 +241,7 @@ def run_benchmark(model, args):
     exe = fluid.Executor(place)
     exe.run(fluid.default_startup_program())
     accuracy = fluid.average.WeightedAverage()
+    train_exe = fluid.ParallelExecutor(use_cuda=True, loss_name=avg_cost.name)
     if args.use_fake_data:
         data = train_reader().next()
         image = np.array(map(lambda x: x[0].reshape(dshape), data)).astype(
@@ -264,14 +265,17 @@ def run_benchmark(model, args):
                                      data)).astype('float32')
                 label = np.array(map(lambda x: x[1], data)).astype('int64')
                 label = label.reshape([-1, 1])
-            loss, acc, weight = exe.run(
-                fluid.default_main_program(),
+            loss, acc, weight = train_exe.run(
                 feed={'data': image,
                       'label': label},
-                fetch_list=[avg_cost, batch_acc, batch_size_tensor])
+                fetch_list=[
+                    avg_cost.name, batch_acc.name, batch_size_tensor.name
+                ])
             iters += 1
             num_samples += len(label)
-            accuracy.add(value=acc, weight=weight)
+            accuracy.add(value=np.array(np.mean(acc)), weight=np.mean(weight))
+            loss = np.mean(np.array(loss))
+            acc = np.mean(np.array(acc))
             train_losses.append(loss)
             train_accs.append(acc)
             print("Pass: %d, Iter: %d, Loss: %f, Accuracy: %f" %

benchmark/fluid/vgg.py

@@ -169,6 +169,7 @@ def main():
 
     iters, num_samples, start_time = 0, 0, time.time()
     accuracy = fluid.average.WeightedAverage()
+    train_exe = fluid.ParallelExecutor(use_cuda=True, loss_name=avg_cost.name)
     for pass_id in range(args.pass_num):
         accuracy.reset()
         train_accs = []
@@ -184,14 +185,17 @@ def main():
             y_data = np.array(map(lambda x: x[1], data)).astype("int64")
             y_data = y_data.reshape([-1, 1])
 
-            loss, acc, weight = exe.run(
-                fluid.default_main_program(),
+            loss, acc, weight = train_exe.run(
                 feed={"pixel": img_data,
                       "label": y_data},
-                fetch_list=[avg_cost, batch_acc, batch_size_tensor])
-            accuracy.add(value=acc, weight=weight)
+                fetch_list=[
+                    avg_cost.name, batch_acc.name, batch_size_tensor.name
+                ])
+            accuracy.add(value=np.array(np.mean(acc)), weight=np.mean(weight))
             iters += 1
             num_samples += len(y_data)
+            loss = np.mean(np.array(loss))
+            acc = np.mean(np.array(acc))
             print(
                 "Pass = %d, Iter = %d, Loss = %f, Accuracy = %f" %
                 (pass_id, iters, loss, acc)

cmake/external/boost.cmake

@@ -24,7 +24,7 @@ set(BOOST_PROJECT       "extern_boost")
 # So we use 1.41.0 here.
 set(BOOST_VER           "1.41.0")
 set(BOOST_TAR           "boost_1_41_0")
-set(BOOST_URL           "http://paddlepaddledeps.bj.bcebos.com/${BOOST_TAR}.tar.gz")
+set(BOOST_URL           "http://paddlepaddledeps.cdn.bcebos.com/${BOOST_TAR}.tar.gz")
 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)

cmake/external/eigen.cmake

@@ -21,11 +21,12 @@ else()
     ExternalProject_Add(
         extern_eigen3
         ${EXTERNAL_PROJECT_LOG_ARGS}
-        GIT_REPOSITORY  "https://github.com/RLovelett/eigen.git"
+        GIT_REPOSITORY  "https://github.com/eigenteam/eigen-git-mirror"
         # eigen on cuda9.1 missing header of math_funtions.hpp
         # https://stackoverflow.com/questions/43113508/math-functions-hpp-not-found-when-using-cuda-with-eigen
         GIT_TAG         917060c364181f33a735dc023818d5a54f60e54c
         PREFIX          ${EIGEN_SOURCE_DIR}
+        DOWNLOAD_NAME   "eigen"
         UPDATE_COMMAND  ""
         CONFIGURE_COMMAND ""
         BUILD_COMMAND     ""

cmake/external/mkldnn.cmake

@@ -45,15 +45,15 @@ IF(${CBLAS_PROVIDER} STREQUAL "MKLML")
 ELSE()
     MESSAGE(FATAL_ERROR "Should enable MKLML when build MKLDNN")
 ENDIF()
-
-SET(MKLDNN_CFLAG "${CMAKE_C_FLAGS} -Wno-error=strict-overflow")
-SET(MKLDNN_CXXFLAG "${CMAKE_CXX_FLAGS} -Wno-error=strict-overflow")
+SET(MKLDNN_FLAG "-Wno-error=strict-overflow -Wno-error=unused-result -Wno-unused-result")
+SET(MKLDNN_CFLAG "${CMAKE_C_FLAGS} ${MKLDNN_FLAG}")
+SET(MKLDNN_CXXFLAG "${CMAKE_CXX_FLAGS} ${MKLDNN_FLAG}")
 ExternalProject_Add(
     ${MKLDNN_PROJECT}
     ${EXTERNAL_PROJECT_LOG_ARGS}
     DEPENDS             ${MKLDNN_DEPENDS}
     GIT_REPOSITORY      "https://github.com/01org/mkl-dnn.git"
-    GIT_TAG             "v0.11"
+    GIT_TAG             "db3424ad44901513c03a1ea31ccaacdf633fbe9f"
     PREFIX              ${MKLDNN_SOURCES_DIR}
     UPDATE_COMMAND      ""
     CMAKE_ARGS          -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL_DIR}
@@ -61,6 +61,7 @@ ExternalProject_Add(
     CMAKE_ARGS          -DMKLROOT=${MKLML_ROOT}
     CMAKE_ARGS          -DCMAKE_C_FLAGS=${MKLDNN_CFLAG}
     CMAKE_ARGS          -DCMAKE_CXX_FLAGS=${MKLDNN_CXXFLAG}
+    CMAKE_ARGS          -DWITH_TEST=OFF -DWITH_EXAMPLE=OFF
     CMAKE_CACHE_ARGS    -DCMAKE_INSTALL_PREFIX:PATH=${MKLDNN_INSTALL_DIR}
                         -DMKLROOT:PATH=${MKLML_ROOT}
 )

cmake/external/mklml.cmake

@@ -27,8 +27,8 @@ ENDIF()
 INCLUDE(ExternalProject)
 
 SET(MKLML_PROJECT       "extern_mklml")
-SET(MKLML_VER           "mklml_lnx_2018.0.1.20171007")
-SET(MKLML_URL           "http://paddlepaddledeps.bj.bcebos.com/${MKLML_VER}.tgz")
+SET(MKLML_VER           "mklml_lnx_2018.0.3.20180406")
+SET(MKLML_URL           "http://paddlepaddledeps.cdn.bcebos.com/${MKLML_VER}.tgz")
 SET(MKLML_SOURCE_DIR    "${THIRD_PARTY_PATH}/mklml")
 SET(MKLML_DOWNLOAD_DIR  "${MKLML_SOURCE_DIR}/src/${MKLML_PROJECT}")
 SET(MKLML_DST_DIR       "mklml")

cmake/external/snappy.cmake

@@ -47,8 +47,6 @@ ExternalProject_Add(
                      -DCMAKE_INSTALL_LIBDIR:PATH=${SNAPPY_INSTALL_DIR}/lib
                      -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
                      -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
-    BUILD_COMMAND   make -j8
-    INSTALL_COMMAND make install
 )
 
 add_library(snappy STATIC IMPORTED GLOBAL)

cmake/external/snappystream.cmake

@@ -46,8 +46,6 @@ ExternalProject_Add(
                         -DCMAKE_INSTALL_PREFIX:PATH=${SNAPPYSTREAM_INSTALL_DIR}
                         -DCMAKE_INSTALL_LIBDIR:PATH=${SNAPPYSTREAM_INSTALL_DIR}/lib
                         -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
-        BUILD_COMMAND   make -j8
-        INSTALL_COMMAND make install
         DEPENDS snappy
 )
 

cmake/inference_lib.cmake

@@ -70,6 +70,12 @@ copy(glog_lib
   DSTS ${dst_dir} ${dst_dir}/lib
 )
 
+set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/boost/")
+copy(boost_lib
+  SRCS ${BOOST_INCLUDE_DIR}/boost
+  DSTS ${dst_dir}
+)
+
 if(NOT PROTOBUF_FOUND)
     set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/protobuf")
     copy(protobuf_lib
@@ -92,6 +98,14 @@ elseif (WITH_MKLML)
     )
 endif()
 
+if(WITH_MKLDNN)
+  set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/mkldnn")
+  copy(mkldnn_lib
+    SRCS ${MKLDNN_INC_DIR} ${MKLDNN_SHARED_LIB}
+    DSTS ${dst_dir} ${dst_dir}/lib
+  )
+endif()
+
 if(NOT MOBILE_INFERENCE AND NOT RPI)
   set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/snappy")
   copy(snappy_lib
@@ -142,4 +156,30 @@ copy(string_lib
   DSTS ${dst_dir}/${module} ${dst_dir}/${module}/tinyformat
 )
 
+set(module "pybind")
+copy(pybind_lib
+  SRCS ${CMAKE_CURRENT_BINARY_DIR}/paddle/fluid/${module}/pybind.h
+  DSTS ${dst_dir}/${module}
+)
+
+# CMakeCache Info
+copy(cmake_cache
+  SRCS ${CMAKE_CURRENT_BINARY_DIR}/CMakeCache.txt
+  DSTS ${CMAKE_INSTALL_PREFIX})
+
 add_custom_target(inference_lib_dist DEPENDS ${inference_lib_dist_dep}) 
+
+# paddle fluid version
+execute_process(
+  COMMAND ${GIT_EXECUTABLE} log --pretty=format:%H -1
+  OUTPUT_VARIABLE PADDLE_GIT_COMMIT)
+set(version_file ${CMAKE_INSTALL_PREFIX}/version.txt)
+file(WRITE ${version_file}
+  "GIT COMMIT ID: ${PADDLE_GIT_COMMIT}\n"
+  "WITH_MKL: ${WITH_MKL}\n"
+  "WITH_GPU: ${WITH_GPU}\n")
+if(WITH_GPU)
+  file(APPEND ${version_file}
+    "CUDA version: ${CUDA_VERSION}\n"
+    "CUDNN version: v${CUDNN_MAJOR_VERSION}\n")
+endif()

contrib/inference/README.md

+# Embed Paddle Inference in Your Application
+
+Paddle inference offers the APIs in `C` and `C++` languages.
+
+One can easily deploy a model trained by Paddle following the steps as below:
+
+1. Optimize the native model;
+2. Write some codes for deployment.
+
+
+Let's explain the steps in detail.
+
+## Optimize the native Fluid Model
+
+The native model that get from the training phase needs to be optimized for that.
+
+- Clean the noise such as the cost operators that do not need inference;
+- Prune unnecessary computation fork that has nothing to do with the output;
+- Remove extraneous variables;
+- Memory reuse for native Fluid executor;
+- Translate the model storage format to some third-party engine's, so that the inference API can utilize the engine for acceleration;
+
+We have an official tool to do the optimization, call `paddle_inference_optimize --help` for more information.
+
+## Write some codes
+
+Read `paddle_inference_api.h` for more information.

contrib/inference/paddle_inference_api.h

+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+   http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License. */
+
+#pragma once
+
+#include <string>
+#include <vector>
+
+namespace paddle {
+
+class Predictor {
+public:
+  struct Attr;
+  Predictor() = default;
+
+  // Build the network before inference.
+  bool Init(const Attr& attr);
+
+  // Predict an record.
+  // Arguments:
+  //   inputs: the name of the input variables.
+  //   outputs: the name of the output varaibles.
+  //   input_shapes: the shape of the input variables.
+  //   output_shapes: the shape of the output variables.
+  //   input_data: the data of the input variables.
+  //   output_data: the data of the output variables.
+  bool Run(const std::vector<std::string>& inputs,
+           const std::vector<std::string>& outputs,
+           const std::vector<std::vector<int>>& input_shapes,
+           const std::vector<std::vector<int>>& output_shapes,
+           const std::vector<std::vector<float>>& input_data,
+           std::vector<std::vector<float>>* output_data);
+
+  // Clone a predictor that share the model weights.
+  Predictor* Clone();
+
+  // Destroy the Predictor.
+  ~Predictor();
+
+  struct Attr {
+    enum class EngineKind;
+
+    std::string model_dir;      // path to the model directory.
+    bool enable_engine{false};  // Enable to execute (part of) the model on
+                                // third-party engines.
+    EngineKind engine_kind{Attr::EngineKind::kNone};
+
+    enum class EngineKind {
+      kNone = -1,          // Use the native Fluid facility.
+      kAnakin,             // Use Anakin for inference.
+      kTensorRT,           // Use TensorRT for inference.
+      kAutoMixedAnakin,    // Automatically mix Fluid with Anakin.
+      kAutoMixedTensorRT,  // Automatically mix Fluid with TensorRT.
+    };
+  };
+};
+
+}  // namespace paddle

doc/fluid/build_and_install/paddleci.png

+../../v2/build_and_install/paddleci.png
\ No newline at end of file

doc/fluid/design/concepts/functions_operators_layers.md

@@ -40,7 +40,7 @@ template <typename T>
 class FCOp : public OperatorBase {
  public:
   void Run(...) {
-    add(mul(Input<T>("X"), Input<T>("W")), Input<T>("b");
+    add(mul(Input<T>("X"), Input<T>("W")), Input<T>("b"));
   }
 };
 REGISTER_OP(FCOp, "fc");

doc/fluid/design/concepts/lod_tensor.md

@@ -155,7 +155,7 @@ into offsets
    3  2+3 4+5 1+9 2+10 3+12
 ```
 
-so we know that the first sentence is from word 0 to word 3, and the second sentence from work 3 to word 5.
+so we know that the first sentence is from word 0 to word 3, and the second sentence from word 3 to word 5.
 
 Similarly, the lengths in the top level LoD
 

doc/fluid/design/dist_train/async_update.md

@@ -4,34 +4,37 @@
 
 For the typical synchronous distributed training, some significant steps are as follows:
 
-1. A Trainer will compute the gradients and SEND them to the Parameter Server(PServer) nodes.
-1. After the PServer node received gradients came from all the Trainers, It will aggregate the
+1. A trainer process will compute the gradients and **send** them to the parameter server (PS) nodes.
+1. After the PS node received gradients came from all the Trainers, It will aggregate the
 gradient variables for the same parameter into one gradient variable and then apply the aggregated
 gradient to the respective parameter, finally using an optimize algorithms(SGD, Monument...)
 to update the parameters.
-1. The Trainer would wait for the PServers finished the optimize stage, and GET the parameters from PServer,
+1. The Trainer would wait for the PS finished the optimize stage, and GET the parameters from PS,
 so all the Trainers would get the same parameters.
 
-In the synchronously distributed training, there should be a `Barrier` to synchronise the
-parameters after the optimizing stage. The performance of a distributed training job would
-depend on the slowest node if there were hundreds or thousands of training nodes in a
-Job, the performance of synchronously distributed training might be very poor because of
-the slow node. So this design doc would introduce an approach to implement
-*asynchronously* distributed training in PaddlePaddle Fluid.
+In Synchronous Distributed Training, there is a **barrier** on each PS to wait until all trainers processes
+have completed running current mini-batch. After that, all trainers can continue to run the next
+mini-batch. So, we can find that the overall performance of Synchronous Distributed Training depends 
+on the slowest node.
+
+In Asynchronous Distributed Training, we don't need to wait for a global mini-bach, the optimizer on
+the PS will run immediately when the gradient is uploaded to the PS from one trainer. This mode would
+train such models that achieve scaling, better throughput. In this design doc, we will introduce how to 
+implement the Asynchronous Distributed Training base on PaddlePaddle Fluid.
 
 ## Design
 
 <img src="./src/async_update.png" width="600"/>
 
-As the figure above, we describe a global view of asynchronously update process and use
+As the figure above, we describe a global view of the asynchronous update process and use
 the parameter `w1` as an example to introduce the steps:
 1. For each gradient variables, they may distribute on different GPU card and aggregate
 them while they are all calculated.
-1. Split the gradient variable into multiple blocks according to the number of PServer
+1. Split the gradient variable into multiple blocks according to the number of PS
 instances and then send them.
-1. PServer would run an `Optimize Block` using a specified optimize algorithm to update
+1. PS would run an `Optimize Block` using a specified optimize algorithm to update
 the specified parameter.
-1. The trainer will fetch latest parameter from PServer before running forward Op which depends
+1. The trainer will fetch the latest parameter from PS before running forward Op which depends
 on the specified parameter.
 1. Broadcast the received variable into multiple GPU cards and continue to run the next
 mini-batch.
@@ -40,8 +43,8 @@ mini-batch.
 
 - For the multiple devices distributed training, we need to aggregate the gradient
 variables which placed on different devices firstly and then schedule a `SendVars` Operator to
-send the gradient variables to the multiple PServer instances.
-- Schedule `FetchVars` operator to fetch the latest parameter from PServer before running
+send the gradient variables to the multiple PS instances.
+- Schedule `FetchVars` operator to fetch the latest parameter from PS before running
 the forward ops.
 - There could be a large number of gradient variables to be sent, so we need to use another
 thread pool(IO Threadpool) whose a number of the schedulable threads is larger than the

doc/fluid/design/dist_train/distributed_traing_review.md

@@ -42,7 +42,3 @@ Codistillation is a technique that tries to scale the training further. A few tr
 [3] Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, et al. Google’s neural machine translation system: Bridging the gap between human and machine translation.
 
 [4] LARGE SCALE DISTRIBUTED NEURAL NETWORK TRAINING THROUGH ONLINE DISTILLATION
-
-
-
-

doc/fluid/design/motivation/api.md

@@ -77,8 +77,7 @@ print "The sematic-vector of testA: ", paddle.infer(fA, parameters, testA)
 
 ### Example 2. Sharing Parameters between "Models"
 
-We use [GAN](https://github.com/PaddlePaddle/book/tree/develop/gan) in
-this example.  In the following example program, `d0` and `d1`
+We use GAN in this example.  In the following example program, `d0` and `d1`
 correspond to the two networks in the following figure:
 
 <img src="https://github.com/wangyang59/book/raw/00036f4b0da5225041a6824587c1a01cf20159b1/gan/image/gan_ig.png" width=400 />

doc/fluid/design/motivation/refactorization.md

@@ -125,12 +125,12 @@ Compile Time -> IR -> Runtime
 
 ## Operator/OpWithKernel/OpKernel
 
-![class_diagram](http://api.paddlepaddle.org/graphviz?dot=https://gist.githubusercontent.com/reyoung/53df507f6749762675dff3e7ce53372f/raw/49caf1fb70820fb4a6c217634317c9306f361f36/op_op_with_kern_class_diagram.dot)
+![class_diagram](https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/fluid/images/op_op_with_kern_class_diagram.dot)
 
 ---
 
 ## Operator
-![class_diagram](http://api.paddlepaddle.org/graphviz?dot=https://gist.githubusercontent.com/reyoung/53df507f6749762675dff3e7ce53372f/raw/dd598e8f1976f5759f58af5e5ef94738a6b2e661/op.dot)
+![class_diagram](https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/fluid/images/op.dot)
 
 * `Operator` is the fundamental building block of the user interface.
     * Operator stores input/output variable names and attributes.
@@ -141,7 +141,7 @@ Compile Time -> IR -> Runtime
 
 ## OpWithKernel/Kernel
 
-![class_diagram](http://api.paddlepaddle.org/graphviz?dot=https://gist.githubusercontent.com/reyoung/53df507f6749762675dff3e7ce53372f/raw/9d7f4eba185cf41c8e2fbfb40ae21890dbddcd39/op_with_kernel.dot)
+![class_diagram](https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/fluid/images/op_with_kernel.dot)
 
 * `OpWithKernel` inherits `Operator`.
 * `OpWithKernel` contains a Kernel map.

doc/fluid/design/multi_devices/operator_kernel_type.md

@@ -75,7 +75,7 @@ Different layout leads to different implementation of the operator kernel. There
 
 - The inference of Layout is at run-time, not at compile-time.
 
-- Every operator has to implement different kernels for different layouts. Let's take MKLDNN as an example. If we want to implement an MKLDNN convolution operator, we have to implement all the kernels for different layouts, which are listed [here](http://01org.github.io/mkl-dnn/structmkldnn_1_1memory.html). And we will have a special macro to  register kernels for MKLDNN operators.
+- Every operator has to implement different kernels for different layouts. Let's take MKLDNN as an example. If we want to implement an MKLDNN convolution operator, we have to implement all the kernels for different layouts, which are listed [here](http://intel.github.io/mkl-dnn/structmkldnn_1_1memory.html). And we will have a special macro to  register kernels for MKLDNN operators.
 
 `Layout` is also defined as a enum variable:
 

doc/fluid/howto/cluster/nccl2_rdma_training.md

+# Distributed Training with NCCL2 and RDMA
+
+When doing distributed multi-GPU training, network bandwith often becomes the
+bottle neck. We introduce a way to use NCCL2 to do such training job to
+achieve best performace.
+
+## Prepare Hardwares with RDMA and Multiple GPUs
+
+I'm using two Linux servers each of them is installed with 8 GPUs and
+one 100Gb RDMA card.
+Base environment is:
+
+* OS: CentOS 7.4
+* RDMA device: "Mellanox Technologies MT27700 Family [ConnectX-4]"
+* Kernel version: `4.4.88-1.el7.elrepo.x86_64`
+* Docker version: `1.12.6`
+* Docker storage driver: `overlay2`
+* IP addresses: 192.168.16.30,192.168.16.34
+
+In general, the steps including:
+
+1. Install GPU drivers
+1. Install RDMA drivers
+1. Install "InfiniBand Support"
+1. Use docker to run tests and make sure GPUs and RDMA can work inside
+   the container.
+
+I'll ommit section "Install GPU drivers" because we can find it easily
+somewhere else.
+
+### Install RDMA drivers
+
+For my case, I've got two machines with device
+"Mellanox Technologies MT27700 Family [ConnectX-4]" installed. The OS was
+"CentOS 7.4" and I updated the kernel to version 4.4 so that docker can
+work with latest overlay2 filesystem.
+
+***NOTE: before you start, make sure you have a way to get a console
+of the server other than ssh because we may need to re-configure the
+network device.***
+
+1. Go to http://www.mellanox.com/page/products_dyn?product_family=26,
+   download `MLNX_OFED` software in the bottom of the page, and upload it
+   onto the server.
+1. Run `./mlnxofedinstall --add-kernel-support` in the software package.
+1. Run `/etc/init.d/openibd restart` to make everything work, note that
+   this operation may cause the network goes down if you are using this
+   RDMA device as default network device and use ssh to login the server.
+1. Re-configure the network interface, for example:
+   `ifconfig eth2 192.168.16.30/20 up`, then add routes if needed:
+   `ip route add default via 192.168.16.1 dev eth2`.
+1. Do the same thing on the other node.
+1. Use `ping` to test if the two nodes have typical ICMP connection.
+1. Use either `udaddy` or `ib_write_bw` to test the network connection is
+   ready and have the desired bandwith.
+
+### Prepare Docker Image to Run RDMA Programs
+
+1. Build a docker image using cuda base image like: `nvidia/cuda:8.0-cudnn5-devel-ubuntu16.04` and install paddlepaddle whl
+   package in it.
+1. Start a docker container and mount GPU driver libs into it (you can
+   skip this step if you are using nvidia-docker).
+1. Mount RDMA dirvers and libs into the docker image (see below section),
+   also `udaddy` and `ib_write_bw` if needed.
+1. Mount GPU devices and RDMA devices into the container using `--device`
+   or just use privileged mode `--privileged`.
+1. Start the container using host network mode: `--net=host`
+
+### RDMA Library Files Needed
+
+Usually, `MLNX_OFED` install latest supported libs under
+`/usr/lib64/mlnx_ofed/valgrind`. Other libs also needed to run RDMA programs
+is listed below. These libs must be mounted into the docker container.
+
+* Libs under `/usr/lib64/mlnx_ofed/valgrind`
+  * libibcm.so
+  * libibverbs.so
+  * libmlx4.so
+  * libmlx5.so
+  * libmlx5-rdmav2.so
+  * librdmacm.so
+* Other libs:
+  * libnl-3.so.200
+  * libnl-route-3.so.200
+  * libnuma.so.1
+
+## Start to Run the Training Job
+
+Setting NCCL environment variables to turn NCCL switches on and off:
+
+
+| Env Name | Description |
+| --- | --- |
+| NCCL_SOCKET_IFNAME | The RDMA device, e.g. eth2 |
+| NCCL_P2P_DISABLE | Set to 1 to disable P2P transfer between GPUs |
+| NCCL_IB_DISABLE | Set to 1 to disable using RDMA |
+| NCCL_IB_CUDA_SUPPORT | Set to 1 to enable GPU Direct if supported |
+| NCCL_DEBUG | Set debug level: VERSION, WARN, INFO |
+
+My two servers are: `192.168.16.30,192.168.16.34`, On node 1, Run :
+
+```bash
+PADDLE_TRAINER_ID=0 PADDLE_PORT=48372 PADDLE_WORKERS=192.168.16.30,192.168.16.34 POD_IP=192.168.16.30 stdbuf -oL python vgg16.py
+```
+
+On node 2, Run:
+
+```bash
+PADDLE_TRAINER_ID=1 PADDLE_PORT=48372 PADDLE_WORKERS=192.168.16.30,192.168.16.34 POD_IP=192.168.16.34 stdbuf -oL python vgg16.py
+```

doc/fluid/howto/optimization/cpu_profiling_cn.md

+# CPU性能调优
+
 此教程会介绍如何使用Python的cProfile包、Python库yep、Google perftools来进行性能分析 (profiling) 与调优（performance tuning）。
 
 Profling 指发现性能瓶颈。系统中的瓶颈可能和程序员开发过程中想象的瓶颈相去甚远。Tuning 指消除瓶颈。性能优化的过程通常是不断重复地 profiling 和 tuning。
@@ -8,7 +10,7 @@ PaddlePaddle 用户一般通过调用 Python API 编写深度学习程序。大
 * Python 与 C++ 混合代码的性能分析
 
 
-# Python代码的性能分析
+## Python代码的性能分析
 
 ### 生成性能分析文件
 

doc/fluid/howto/optimization/cpu_profiling_en.md

+# Tune CPU performance
+
 This tutorial introduces techniques we use to profile and tune the
 CPU performance of PaddlePaddle.  We will use Python packages
 `cProfile` and `yep`, and Google's `perftools`.
@@ -14,7 +16,7 @@ the profiling and tuning of
 1. the Python code and
 1. the mixture of Python and C++ code.
 
-# Profiling the Python Code
+## Profiling the Python Code
 
 ### Generate the Performance Profiling File
 

doc/fluid/images/op.dot

+digraph sample { 
+  graph [rankdir=TD]; node [shape=record];
+  op [label="{Operator| InferShape()=0\lRun()=0\l | map<string, string[]> inputs_\lmap<string, string[]> outputs_ \l AttributeMap attrs_\l}"]; 
+}
\ No newline at end of file

doc/fluid/images/op_op_with_kern_class_diagram.dot

+digraph sample { 
+  graph [rankdir=TD]; node [shape=record];
+  op [label="{Operator| InferShape()=0\lRun()=0\l | map<string, string[]> inputs_\lmap<string, string[]> outputs_ \l AttributeMap attrs_\l}"]; 
+  op_with_kern [label="{OpWithKernel | InferShape()=0\lRun()\l | map<OpKernelKey,OpKernel>kernels_ }"]
+  op_kernel [label="{OpKernel | Compute()=0}"]
+  op_kernel_key [label="{OpKernelKey| Place place\n...}"]
+
+  op -> op_with_kern [dir=back, arrowtail=onormal]
+  op_with_kern -> op_kernel [arrowhead=vee, label="contains many"]
+
+  {
+    rank=same;
+    op_with_kern
+    op_kernel
+  }
+
+  op_kernel -> op_kernel_key [style=invis]
+
+  {
+    rank=same;
+    op_kernel
+    op_kernel_key
+  }
+
+  op_with_kern -> op_kernel_key [arrowhead=vee, label ="\nas map key"]
+
+  mul_op [label="MulOp"]
+  op_with_kern -> mul_op [dir=back, arrowtail=onormal]
+  mul_kernel [label="template <typename Place>\lclass MulOpKernel\l"]
+  op_kernel -> mul_kernel [dir=back, arrowtail=onormal]
+  mul_op -> mul_kernel [arrowhead=vee, label="register many"]
+  
+  {
+    rank=same;
+    mul_op;
+    mul_kernel;
+  }
+}
\ No newline at end of file

doc/fluid/images/op_with_kernel.dot

+digraph sample { 
+  graph [rankdir=TD]; node [shape=record];
+  op [label="{Operator}"]; 
+  op_with_kern [label="{OpWithKernel | InferShape()=0\lRun()\l | map<OpKernelKey,OpKernel>kernels_ }"]
+  op_kernel [label="{OpKernel | Compute()=0}"]
+  op_kernel_key [label="{OpKernelKey| Place place\n...}"]
+
+  op -> op_with_kern [dir=back, arrowtail=onormal]
+  op_with_kern -> op_kernel [arrowhead=vee, label="contains many"]
+
+  {
+    rank=same;
+    op_with_kern
+    op_kernel
+  }
+
+  op_kernel -> op_kernel_key [style=invis]
+
+  {
+    rank=same;
+    op_kernel
+    op_kernel_key
+  }
+
+  op_with_kern -> op_kernel_key [arrowhead=vee, label ="\nas map key"]
+}
\ No newline at end of file

doc/v2/api/config/layer.rst

@@ -142,7 +142,7 @@ gated_unit
 -----------
 ..  autoclass:: paddle.v2.layer.gated_unit
     :noindex:
-    
+
 Recurrent Layer Group
 =====================
 
@@ -354,7 +354,7 @@ dropout
 --------
 ..  autoclass:: paddle.v2.layer.dropout
     :noindex:
-    
+
 dot_prod
 ---------
 .. autoclass:: paddle.v2.layer.dot_prod
@@ -460,6 +460,11 @@ multi_binary_label_cross_entropy_cost
 ..  autoclass:: paddle.v2.layer.multi_binary_label_cross_entropy_cost
     :noindex:
 
+classification_cost
+-------------------
+.. autoclass:: paddle.v2.layer.classification_cost
+   :noindex:
+
 huber_regression_cost
 -------------------------
 ..  autoclass:: paddle.v2.layer.huber_regression_cost
@@ -534,7 +539,7 @@ detection_output
 ----------------
 ..  autoclass:: paddle.v2.layer.detection_output
     :noindex:
-    
+
 Check Layer
 ============
 

doc/v2/build_and_install/build_from_source_cn.rst

@@ -19,8 +19,9 @@
 ----------------
 
 PaddlePaddle需要使用Docker环境完成编译，这样可以免去单独安装编译依赖的步骤，可选的不同编译环境Docker镜像
-可以在 `这里 <https://hub.docker.com/r/paddlepaddle/paddle_manylinux_devel/tags/>`_ 找到。或者
-参考下述可选步骤，从源码中构建用于编译PaddlePaddle的Docker镜像。
+可以在 `这里 <https://hub.docker.com/r/paddlepaddle/paddle_manylinux_devel/tags/>`_ 找到，您也可以
+在 `这里 <https://github.com/PaddlePaddle/Paddle/tree/develop/tools/manylinux1/>`_ 找到 paddle_manylinux_devel
+镜像的编译以及使用方法。或者参考下述可选步骤，从源码中构建用于编译PaddlePaddle的Docker镜像。
 
 如果您选择不使用Docker镜像，则需要在本机安装下面章节列出的 `编译依赖`_ 之后才能开始编译的步骤。
 

doc/v2/build_and_install/build_from_source_en.rst

@@ -22,6 +22,8 @@ How To Build
 You need to use Docker to build PaddlePaddle
 to avoid installing dependencies by yourself. We have several pre-built
 Docker images `here <https://hub.docker.com/r/paddlepaddle/paddle_manylinux_devel/tags/>`_ ,
+you can also find how to build and use paddle_manylinux_devel Docker image from
+`here <https://github.com/PaddlePaddle/Paddle/tree/develop/tools/manylinux1/>`_
 Or you can build your own image from source as the optional step below:
 
 .. code-block:: bash

doc/v2/build_and_install/pip_install_cn.rst

@@ -10,20 +10,38 @@ PaddlePaddle可以使用常用的Python包管理工具
 使用pip安装
 ------------------------------
 
-
-执行下面的命令即可在当前机器上安装PaddlePaddle的运行时环境，并自动下载安装依赖软件，版本为cpu_avx_openblas。
+执行下面的命令即可在当前机器上安装PaddlePaddle的运行时环境，并自动下载安装依赖软件。
 
   .. code-block:: bash
 
      pip install paddlepaddle
 
+当前的默认版本为0.12.0，cpu_avx_openblas，您可以通过指定版本号来安装其它版本，例如:
+
+  .. code-block:: bash
+
+      pip install paddlepaddle==0.11.0
+
 
-如果需要安装支持GPU的版本（cuda7.5_cudnn5_avx_openblas），需要执行：
+如果需要安装支持GPU的版本（cuda8.0_cudnn5_avx_openblas），需要执行：
 
   .. code-block:: bash
 
      pip install paddlepaddle-gpu
 
+当前的默认版本也是0.12.0，PaddlePaddle针对不同需求提供了更多版本的安装包，部分列表如下：
+
+=================================   ========================================
+版本号                               版本说明
+=================================   ========================================
+paddlepaddle-gpu==0.12.0            使用CUDA 8.0和cuDNN 5编译的0.12.0版本
+paddlepaddle-gpu==0.11.0.post87     使用CUDA 8.0和cuDNN 7编译的0.11.0版本
+paddlepaddle-gpu==0.11.0.post8      使用CUDA 8.0和cuDNN 5编译的0.11.0版本
+paddlepaddle-gpu==0.11.0            使用CUDA 7.5和cuDNN 5编译的0.11.0版本
+=================================   ========================================
+
+您可以在 `Release History <https://pypi.org/project/paddlepaddle-gpu/#history>`_ 中找到paddlepaddle-gpu的各个发行版本。
+
 如果需要获取并安装最新的（开发分支）PaddlePaddle，可以从我们的CI系统中下载最新的whl安装包和c-api开发包并安装，
 您可以从下面的表格中找到需要的版本：
 
@@ -37,12 +55,11 @@ PaddlePaddle可以使用常用的Python包管理工具
     :header: "版本说明", "cp27-cp27mu", "cp27-cp27m"
     :widths: 1, 3, 3
 
-    "cpu_avx_mkl", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cpu_avx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cpu_noavx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cuda7.5_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
+    "cpu_avx_mkl", "`paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl>`_"
+    "cpu_avx_openblas", "`paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl>`_"
+    "cpu_noavx_openblas", "`paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl>`_"
+    "cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`_"
+    "cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`_"
 
 .. _pip_dependency:
 
@@ -69,7 +86,7 @@ PaddlePaddle发布的安装包会尽量对齐 `manylinux1 <https://www.python.or
 ------------------------------
 
 - paddlepaddle*.whl is not a supported wheel on this platform.
-  
+
   出现这个问题的主要原因是，没有找到和当前系统匹配的paddlepaddle安装包。请检查Python版本是否为2.7系列。另外最新的pip官方源中的安装包默认是manylinux1标准，需要使用最新的pip (>9.0.0) 才可以安装。可以使用下面的命令更新您的pip：
 
     .. code-block:: bash

doc/v2/build_and_install/pip_install_en.rst

@@ -12,20 +12,38 @@ Install using pip
 ------------------------------
 
 Run the following command to install PaddlePaddle on the current
-machine, it will also download requirements, the version is cpu_avx_openblas.
+machine, it will also download requirements.
 
   .. code-block:: bash
 
      pip install paddlepaddle
 
+the default version is 0.12.0, cpu_avx_openblas, you can specify the versions to satisfy your demands, like:
 
-If you wish to install GPU version (cuda7.5_cudnn5_avx_openblas), just run:
+  .. code-block:: bash
+
+      pip install paddlepaddle==0.11.0
+
+If you need to install a GPU-enabled version (cuda8.0_cudnn5_avx_openblas), you need to run:
 
   .. code-block:: bash
 
      pip install paddlepaddle-gpu
 
-If you wish to install the latest develop branch PaddlePaddle, 
+The default version is also 0.12.0, PaddlePaddle provides several versions of packages for different needs, as shown in the table:
+
+=================================   ========================================
+版本号                               版本说明
+=================================   ========================================
+paddlepaddle-gpu==0.12.0            0.12.0 built with CUDA 8.0 and cuDNN 5
+paddlepaddle-gpu==0.11.0.post87     0.11.0 built with CUDA 8.0 and cuDNN 7
+paddlepaddle-gpu==0.11.0.post8      0.11.0 built with CUDA 8.0 and cuDNN 5
+paddlepaddle-gpu==0.11.0            0.11.0 built with CUDA 7.5 and cuDNN 5
+=================================   ========================================
+
+You can find all versions released of paddlepaddle-gpu in `Release History <https://pypi.org/project/paddlepaddle-gpu/#history>`_ .
+
+If you wish to install the latest develop branch PaddlePaddle,
 you can download the latest whl package from our CI system. Access
 the below links, log in as guest, then click at the "Artifact"
 tab, you'll find the download link of whl packages.
@@ -40,12 +58,11 @@ If the links below shows up the login form, just click "Log in as guest" to star
     :header: "version", "cp27-cp27mu", "cp27-cp27m"
     :widths: 1, 3, 3
 
-    "cpu_avx_mkl", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cpu_avx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cpu_noavx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cuda7.5_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda75cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
-    "cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl>`_"
+    "cpu_avx_mkl", "`paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxCp27cp27mu/.lastSuccessful/paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl>`_"
+    "cpu_avx_openblas", "`paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuAvxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl>`_"
+    "cpu_noavx_openblas", "`paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_CpuNoavxOpenblas/.lastSuccessful/paddlepaddle-latest-cp27-cp27m-linux_x86_64.whl>`_"
+    "cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda80cudnn5cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`_"
+    "cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27mu-linux_x86_64.whl>`_", "`paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl <https://guest:@paddleci.ngrok.io/repository/download/Manylinux1_Cuda8cudnn7cp27cp27mu/.lastSuccessful/paddlepaddle_gpu-latest-cp27-cp27m-linux_x86_64.whl>`_"
 
 .. _pip_dependency:
 
@@ -79,7 +96,7 @@ FAQ
 ------------------------------
 
 - paddlepaddle*.whl is not a supported wheel on this platform.
-  
+
   The main cause of this issue is that your current platform is
   not supported. Please check that you are using Python 2.7 series.
   Besides, pypi only supports manylinux1 standard, you'll need to

doc/v2/design/mkl/mkldnn.md

@@ -5,7 +5,7 @@
 充分展现英特尔平台的优势，有效提升PaddlePaddle在英特尔架构上的性能。
 
 <div align="center">
-<img src="image/overview.png"><br/>
+<img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/v2/images/overview.png"><br/>
 Figure 1. PaddlePaddle on IA
 </div>
 
@@ -42,16 +42,43 @@ Figure 1. PaddlePaddle on IA
 
 MKL，MKLML以及MKL-DNN三者关系如下表：
 
-| Name        |  Open Source     | License     | Descriptions  |
-| :---------- | :--------------- | :---------- | :------------ |
-|   MKL       |     No           | Proprietary | Accelerate math processing routines |
-|   MKLML     |     No           | Proprietary | Small package of MKL, especially for Machine Learning |
-|   MKL-DNN   |     Yes          | Apache 2.0  | Accelerate primitives processing routines especially for Deep Neural Networks  |
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Open Source</th>
+<th>License</th>
+<th>Descriptions</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>MKL</td>
+<td>No</td>
+<td>Proprietary</td>
+<td>Accelerate math processing routines</td>
+</tr>
+<tr>
+<td>MKLML</td>
+<td>No</td>
+<td>Proprietary</td>
+<td>Small package of MKL, especially for Machine Learning</td>
+</tr>
+
+<tr>
+<td>MKL-DNN</td>
+<td>Yes</td>
+<td>Apache 2.0</td>
+<td>Accelerate primitives processing routines especially for Deep Neural Networks</td>
+</tr>
+
+</tbody>
+</table>
 
 MKLML可以与MKL-DNN共同使用，以此达到最好的性能。
 
 <div align="center">
-<img src="image/engine.png"><br/>
+<img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/v2/images/engine.png"><br/>
 Figure 2. PaddlePaddle with MKL Engines
 </div>
 
@@ -103,7 +130,7 @@ MKL-DNN的库目前只有动态库`libmkldnn.so`。
 所以我们定义了一个`MKLDNNMatrix`用于管理MKL-DNN数据的不同格式以及相互之间的转换。
 
 <div align="center">
-<img src="image/matrix.png"><br/>
+<img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/v2/images/matrix.png"><br/>
 Figure 3. MKLDNNMatrix
 </div>
 
@@ -113,7 +140,7 @@ Figure 3. MKLDNNMatrix
 子类只需要使用定义好的接口，实现具体的函数功能即可。
 
 <div align="center">
-<img src="image/layers.png"><br/>
+<img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/v2/images/layers.png"><br/>
 Figure 4. MKLDNNLayer
 </div>
 
@@ -150,7 +177,7 @@ Figure 4. MKLDNNLayer
 所以整体上，在实现每个子类的时候就不需要关心分支的事情了。
 
 <div align="center">
-<img src="image/gradients.png"><br/>
+<img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/v2/images/gradients.png"><br/>
 Figure 5. Merge Gradients
 </div>
 

doc/v2/howto/cluster/multi_cluster/k8s_distributed_en.md

@@ -41,7 +41,7 @@ Training docker image needs to package the paddle pserver and paddle trainer run
 - Generating the initialization arguments for `Paddle PServer` and `Paddle Training` processes.
 
 Since the paddlepaddle official docker image already has the runtimes we need, we'll take it as the base image and pack some additional scripts for the processes mentioned above to build our training image. for more detail, please find from the following link:
-- https://github.com/PaddlePaddle/Paddle/blob/develop/doc/howto/usage/cluster/src/k8s_train/Dockerfile
+- https://github.com/PaddlePaddle/Paddle/tree/develop/doc/v2/howto/cluster/multi_cluster/src/k8s_train/Dockerfile
 
 
 ```bash
@@ -62,7 +62,7 @@ represent the Docker Image which built in this step.
 ### Prepare Training Data
 
 We can download and split the training job by creating a Kubernetes Job, or custom your image
-by editing [k8s_train](./src/k8s_train/).
+by editing [k8s_train](https://github.com/PaddlePaddle/Paddle/tree/develop/doc/v2/howto/cluster/multi_cluster/src/k8s_train).
 
 Before creating a Job, we need to bind a [persistenVolumeClaim](https://kubernetes.io/docs/user-guide/persistent-volumes) by the different type of
 the different file system, the generated dataset would be saved on this volume.

doc/v2/images/glossary_rnn.dot

+digraph G{
+	subgraph cluster_timestep0 {
+		label="recurrent timestep i-1"
+		bgcolor=lightgray
+		node [style=filled,color=white]
+		fc0_0 [label="fc 0"]
+		fc0_1 [label="fc 1"]
+		fc0_2 [label="fc 2"]
+
+		fc0_0 -> fc0_1
+		fc0_1 -> fc0_2
+	}
+
+	subgraph cluster_timestep1 {
+		label="recurrent timestep i"
+		node [style=filled];
+		fc1_0 [label="fc 0"]
+		fc1_1 [label="fc 1"]
+		fc1_2 [label="fc 2"]
+		color=blue
+
+		fc1_0 -> fc1_1
+		fc1_1 -> fc1_2
+	}
+
+	subgraph cluster_timestep2 {
+		label="recurrent timestep i+1"
+		bgcolor=lightgray
+		node [style=filled,color=white]
+		fc2_0 [label="fc 0"]
+		fc2_1 [label="fc 1"]
+		fc2_2 [label="fc 2"]
+
+		fc2_0 -> fc2_1
+		fc2_1 -> fc2_2
+	}
+	
+	
+	fc0_1 -> fc1_1 [style="dotted" constraint=false]
+	fc1_1 -> fc2_1 [style="dotted" constraint=false]
+
+}
\ No newline at end of file

doc/v2/images/glossary_rnn_with_memory.dot

+digraph G{
+	subgraph cluster_timestep0 {
+		label="recurrent timestep i-1"
+		bgcolor=lightgray
+		node [style=filled,color=white]
+		fc0_0 [label="fc 0"]
+		fc0_1 [label="fc 1"]
+		fc0_2 [label="fc 2"]
+		m0 [label="memory"]
+		fc0_0 -> fc0_1
+		fc0_1 -> fc0_2
+		fc0_1 -> m0
+		m0 -> fc0_1
+	}
+
+	subgraph cluster_timestep1 {
+		label="recurrent timestep i"
+		node [style=filled];
+		fc1_0 [label="fc 0"]
+		fc1_1 [label="fc 1"]
+		fc1_2 [label="fc 2"]
+		m1 [label="memory"]
+		color=blue
+		fc1_0 -> fc1_1
+		fc1_1 -> fc1_2
+		fc1_1 -> m1
+		m1 -> fc1_1
+	}
+
+	subgraph cluster_timestep2 {
+		label="recurrent timestep i+1"
+		bgcolor=lightgray
+		node [style=filled,color=white]
+		fc2_0 [label="fc 0"]
+		fc2_1 [label="fc 1"]
+		fc2_2 [label="fc 2"]
+		m2 [label="memory"]
+		fc2_0 -> fc2_1
+		fc2_1 -> fc2_2
+		fc2_1 -> m2
+		m2 -> fc2_1
+	}
+	
+	
+	m0 -> m1 [style="dotted" constraint=false]
+	m1 -> m2 [style="dotted" constraint=false]
+
+}
\ No newline at end of file

doc/v2/images/simple_full_hierarchical_recurrent.dot

+digraph G {
+  rankdir=LR;
+
+  subgraph cluster_t0 {
+    a [label="4"]
+    b [label="5"]
+    c [label="2"]
+  }
+  
+  subgraph cluster_t1 {
+    d [label="0"]
+    e [label="9"]
+  }
+
+  subgraph cluster_t2 {
+    f [label="8"]
+    g [label="1"]
+    h [label="4"]
+  }
+
+  a -> b;
+  b -> c;
+  c -> d [constraint=false];
+
+  d -> e;
+  e -> f [constraint=false];
+  
+  f -> g;
+  g -> h;
+}
\ No newline at end of file

doc/v2/images/simple_full_recurrent.dot

+digraph G {
+  rankdir=LR;
+  a [label="4"]
+  b [label="5"]
+  c [label="2"]
+  d [label="0"]
+  e [label="9"]
+  f [label="8"]
+  g [label="1"]
+  h [label="4"]
+
+  a -> b;
+  b -> c;
+  c -> d;
+  d -> e;
+  e -> f;
+  f -> g;
+  g -> h;
+}
\ No newline at end of file

paddle/CMakeLists.txt

@@ -24,6 +24,6 @@ if(NOT WITH_FLUID_ONLY)
 endif()
 
 add_subdirectory(testing)
-if(NOT MOBILE_INFERENCE AND NOT RPI)
+if(NOT MOBILE_INFERENCE AND NOT RPI AND NOT WITH_C_API)
   add_subdirectory(fluid)
 endif()

paddle/fluid/framework/CMakeLists.txt

@@ -5,11 +5,11 @@ proto_library(framework_proto SRCS framework.proto)
 cc_library(ddim SRCS ddim.cc DEPS eigen3 boost)
 cc_test(ddim_test SRCS ddim_test.cc DEPS ddim)
 nv_test(dim_test SRCS dim_test.cu DEPS ddim)
-
+cc_library(data_type SRCS data_type.cc DEPS framework_proto ddim device_context)
 if(WITH_GPU)
-  nv_library(tensor SRCS tensor.cc tensor_util.cu DEPS ddim place memory device_context framework_proto)
+  nv_library(tensor SRCS tensor.cc tensor_util.cu DEPS place memory data_type)
 else()
-  cc_library(tensor SRCS tensor.cc tensor_util.cc DEPS ddim place memory device_context framework_proto)
+  cc_library(tensor SRCS tensor.cc tensor_util.cc DEPS place memory data_type)
 endif()
 
 cc_test(tensor_test SRCS tensor_test.cc DEPS tensor)
@@ -57,7 +57,7 @@ cc_library(data_transform SRCS data_transform.cc DEPS math_function tensor
 cc_library(attribute SRCS attribute.cc DEPS framework_proto boost)
 cc_test(program_desc_test SRCS program_desc_test.cc DEPS proto_desc
 device_context)
-cc_library(op_proto_maker SRCS op_proto_maker.cc DEPS framework_proto attribute)
+cc_library(op_proto_maker SRCS op_proto_maker.cc DEPS framework_proto attribute glog)
 cc_test(op_proto_maker_test SRCS op_proto_maker_test.cc DEPS op_proto_maker)
 cc_library(op_info SRCS op_info.cc DEPS attribute framework_proto)
 cc_library(shape_inference SRCS shape_inference.cc DEPS ddim attribute device_context)

paddle/fluid/framework/block_desc.cc

@@ -134,6 +134,11 @@ OpDesc *BlockDesc::PrependOp() {
   return ops_.front().get();
 }
 
+void BlockDesc::PrependAllocatedOp(std::unique_ptr<OpDesc> &&op_desc) {
+  need_update_ = true;
+  ops_.emplace_front(std::move(op_desc));
+}
+
 OpDesc *BlockDesc::InsertOp(size_t index) {
   need_update_ = true;
   auto it = ops_.begin() + index;
@@ -143,7 +148,7 @@ OpDesc *BlockDesc::InsertOp(size_t index) {
 }
 
 void BlockDesc::RemoveOp(size_t s, size_t e) {
-  if (ops_.begin() + s == ops_.end() || ops_.begin() + e == ops_.end()) {
+  if (ops_.begin() + s >= ops_.end() || ops_.begin() + e > ops_.end()) {
     return;
   }
   need_update_ = true;