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

Author: JiayiFeng

.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 @@ def str2bool(v):
     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 train_loop(exe, trainer_prog):
             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 train_loop(exe, trainer_prog):
                     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 test(exe):
     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 test(exe):
                                      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 test(exe):
 
     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 test(exe):
             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     ""