Skip to content

P
Paddle

Crayon鑫 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 06c7686a 编写于 作者: R ranqiu
浏览文件
操作

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

上级 7461b359 374e1685
展开全部 隐藏空白更改
内联 并排
Showing with 47320 addition and 7918 deletion
.gitignore
浏览文件 @ 06c7686a
......@@ -21,10 +21,11 @@ third_party/
cmake-build-*
# generated while compiling
python/paddle/v2/framework/core.so
python/paddle/v2/fluid/core.so
paddle/pybind/pybind.h
CMakeFiles
cmake_install.cmake
paddle/.timestamp
python/paddlepaddle.egg-info/
paddle/pybind/pybind.h
python/paddle/v2/framework/tests/tmp/*
.travis.yml
浏览文件 @ 06c7686a
......@@ -30,6 +30,7 @@ addons:
- automake
- libtool
- ccache
ssh_known_hosts: 52.76.173.135
before_install:
- if [[ "$JOB" == "check_style" ]]; then sudo ln -s /usr/bin/clang-format-3.8 /usr/bin/clang-format; fi
# Paddle is using protobuf 3.1 currently. Protobuf 3.2 breaks the compatibility. So we specify the python
......@@ -42,6 +43,14 @@ script:
- |
timeout 2580 paddle/scripts/travis/${JOB}.sh # 43min timeout
RESULT=$?; if [ $RESULT -eq 0 ] || [ $RESULT -eq 142 ]; then true; else false; fi;
- |
if [[ "$JOB" != "build_doc" ]]; then exit 0; fi;
if [[ "$TRAVIS_PULL_REQUEST" != "false" ]]; then exit 0; fi;
if [[ "$TRAVIS_BRANCH" != "develop" && ! "$TRAVIS_BRANCH" =~ ^v[[:digit:]]+\.[[:digit:]]+(\.[[:digit:]]+)?(-\S*)?$ ]]; then exit 0; fi;
export DEPLOY_DOCS_SH=https://raw.githubusercontent.com/PaddlePaddle/PaddlePaddle.org/master/scripts/deploy/deploy_docs.sh
export DOCS_DIR=`pwd`
cd ..
curl $DEPLOY_DOCS_SH | bash -s $CONTENT_DEC_PASSWD $TRAVIS_BRANCH $DOCS_DIR $DOCS_DIR/build/doc
notifications:
email:
on_success: change
......
CMakeLists.txt
浏览文件 @ 06c7686a
......@@ -126,7 +126,8 @@ include(external/swig) # download, build, install swig
include(external/warpctc) # download, build, install warpctc
include(external/any) # download libn::any
include(external/eigen) # download eigen3
include(external/pybind11) # download pybind11
include(external/pybind11) # download pybind11
include(external/nccl)
include(cudnn) # set cudnn libraries, must before configure
include(configure) # add paddle env configuration
......@@ -159,7 +160,7 @@ set(EXTERNAL_LIBS
if(WITH_GPU)
list(APPEND EXTERNAL_LIBS ${CUDA_LIBRARIES} ${CUDA_rt_LIBRARY})
if(NOT WITH_DSO)
list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY})
list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY} ${NCCL_LIBRARY})
endif(NOT WITH_DSO)
endif(WITH_GPU)
......
CONTRIBUTING.md
浏览文件 @ 06c7686a
./doc/howto/dev/contribute_to_paddle_en.md
# Contribute Code
We sincerely appreciate your contribution. This document explains our workflow and work style.
## Workflow
PaddlePaddle uses this [Git branching model](http://nvie.com/posts/a-successful-git-branching-model/). The following steps guide usual contributions.
1. Fork
Our development community has been growing fastly; it doesn't make sense for everyone to write into the official repo. So, please file Pull Requests from your fork. To make a fork, just head over to the GitHub page and click the ["Fork" button](https://help.github.com/articles/fork-a-repo/).
1. Clone
To make a copy of your fork to your local computers, please run
```bash
git clone https://github.com/your-github-account/paddle
cd paddle
```
1. Create the local feature branch
For daily works like adding a new feature or fixing a bug, please open your feature branch before coding:
```bash
git checkout -b my-cool-stuff
```
1. Commit
Before issuing your first `git commit` command, please install [`pre-commit`](http://pre-commit.com/) by running the following commands:
```bash
pip install pre-commit
pre-commit install
```
Our pre-commit configuration requires clang-format 3.8 for auto-formating C/C++ code and yapf for Python.
Once installed, `pre-commit` checks the style of code and documentation in every commit. We will see something like the following when you run `git commit`:
```
➜ git commit
CRLF end-lines remover...............................(no files to check)Skipped
yapf.................................................(no files to check)Skipped
Check for added large files..............................................Passed
Check for merge conflicts................................................Passed
Check for broken symlinks................................................Passed
Detect Private Key...................................(no files to check)Skipped
Fix End of Files.....................................(no files to check)Skipped
clang-formater.......................................(no files to check)Skipped
[my-cool-stuff c703c041] add test file
1 file changed, 0 insertions(+), 0 deletions(-)
create mode 100644 233
```
1. Build and test
Users can build PaddlePaddle natively on Linux and Mac OS X. But to unify the building environment and to make it easy for debugging, the recommended way is [using Docker](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/howto/dev/build_en.md).
1. Keep pulling
An experienced Git user pulls from the official repo often -- daily or even hourly, so they notice conflicts with others work early, and it's easier to resolve smaller conflicts.
```bash
git remote add upstream https://github.com/PaddlePaddle/Paddle
git pull upstream develop
```
1. Push and file a pull request
You can "push" your local work into your forked repo:
```bash
git push origin my-cool-stuff
```
The push allows you to create a pull request, requesting owners of this [official repo](https://github.com/PaddlePaddle/Paddle) to pull your change into the official one.
To create a pull request, please follow [these steps](https://help.github.com/articles/creating-a-pull-request/).
If your change is for fixing an issue, please write ["Fixes <issue-URL>"](https://help.github.com/articles/closing-issues-using-keywords/) in the description section of your pull request. Github would close the issue when the owners merge your pull request.
Please remember to specify some reviewers for your pull request. If you don't know who are the right ones, please follow Github's recommendation.
1. Delete local and remote branches
To keep your local workspace and your fork clean, you might want to remove merged branches:
```bash
git push origin :my-cool-stuff
git checkout develop
git pull upstream develop
git branch -d my-cool-stuff
```
### Code Review
- Please feel free to ping your reviewers by sending them the URL of your pull request via IM or email. Please do this after your pull request passes the CI.
- Please answer reviewers' every comment. If you are to follow the comment, please write "Done"; please give a reason otherwise.
- If you don't want your reviewers to get overwhelmed by email notifications, you might reply their comments by [in a batch](https://help.github.com/articles/reviewing-proposed-changes-in-a-pull-request/).
- Reduce the unnecessary commits. Some developers commit often. It is recommended to append a sequence of small changes into one commit by running `git commit --amend` instead of `git commit`.
## Coding Standard
### Code Style
Our C/C++ code follows the [Google style guide](http://google.github.io/styleguide/cppguide.html).
Our Python code follows the [PEP8 style guide](https://www.python.org/dev/peps/pep-0008/).
Our build process helps to check the code style. In [`build.sh`](https://github.com/PaddlePaddle/Paddle/blob/b84e8226514b8bb4405c3c28e54aa5077193d179/paddle/scripts/docker/build.sh#L42), the entry point of our [builder Docker image](https://github.com/PaddlePaddle/Paddle/blob/b84e8226514b8bb4405c3c28e54aa5077193d179/Dockerfile#L88), the CMake argument `WITH_STYLE_CHECK` is set to `ON` by default. This flag is on
Please install pre-commit, which automatically reformat the changes to C/C++ and Python code whenever we run `git commit`. To check the whole codebase, we can run the command `pre-commit run -a`, as in the [`check_style.sh` file](https://github.com/PaddlePaddle/Paddle/blob/b84e8226514b8bb4405c3c28e54aa5077193d179/paddle/scripts/travis/check_style.sh#L30), which is invoked by [our Travis CI configuration](https://github.com/PaddlePaddle/Paddle/blob/b84e8226514b8bb4405c3c28e54aa5077193d179/.travis.yml#L43).
### Unit Tests
Please remember to add related unit tests.
- For C/C++ code, please follow [`google-test` Primer](https://github.com/google/googletest/blob/master/googletest/docs/Primer.md).
- For Python code, please use [Python's standard `unittest` package](http://pythontesting.net/framework/unittest/unittest-introduction/).
### Writing Logs
We use [glog](https://github.com/google/glog) for logging in our C/C++ code.
For general information, please use `LOG`. For debug information, please use [`VLOG`](http://htmlpreview.github.io/?https://github.com/google/glog/blob/master/doc/glog.html#verbose). The reason is at [here](https://groups.google.com/a/chromium.org/d/msg/chromium-dev/3NDNd1KzXeY/AZKMMx37fdQJ).
`VLOG` requires a *verbose level* parameter. For example:
```c++
VLOG(3) << "Operator FC is taking " << num_inputs << "inputs."
```
When we run a PaddlePaddle application or test, we can specify a verbose threshold. For example:
```bash
GLOG_vmodule=buddy_allocator=2 \
GLOG_v=10 \
python \
../python/paddle/v2/framework/tests/test_recurrent_op.py
```
This will enable VLOG messages generated by `buddy_allocator.{h,cc}` and in the verbose range of 0 to 3, so you will see above example VLOG message, which is in level 3. This suggests that we output overall messages in lower verbose levels, so they display with higher probability. When coding C++, please follow the verbose level convention as follows:
- verbose level 1: [framework](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/framework)
- verbose level 3: [operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators)
- verbose level 5: [memory](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/memory), [platform](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/platform)
- verbose level 7: [math](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/math)
Dockerfile
浏览文件 @ 06c7686a
......@@ -22,7 +22,7 @@ COPY ./paddle/scripts/docker/root/ /root/
RUN apt-get update && \
apt-get install -y \
git python-pip python-dev openssh-server bison \
git python-pip python-dev openssh-server bison libnccl-dev \
wget unzip unrar tar xz-utils bzip2 gzip coreutils ntp \
curl sed grep graphviz libjpeg-dev zlib1g-dev \
python-matplotlib gcc-4.8 g++-4.8 \
......
benchmark/IntelOptimizedPaddle.md 0 → 100644
浏览文件 @ 06c7686a
# Benchmark
Machine:
- Server
- Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz, 2 Sockets, 20 Cores per socket
- Laptop
- DELL XPS15-9560-R1745: i7-7700HQ 8G 256GSSD
- i5 MacBook Pro (Retina, 13-inch, Early 2015)
- Desktop
- i7-6700k
System: CentOS release 6.3 (Final), Docker 1.12.1.
PaddlePaddle: paddlepaddle/paddle:latest (TODO: will rerun after 0.11.0)
- MKL-DNN tag v0.10
- MKLML 2018.0.20170720
- OpenBLAS v0.2.20
On each machine, we will test and compare the performance of training on single node using MKL-DNN / MKLML / OpenBLAS respectively.
## Benchmark Model
### Server
Test on batch size 64, 128, 256 on Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz
Input image size - 3 * 224 * 224, Time: images/second
- VGG-19
| BatchSize | 64 | 128 | 256 |
|--------------|-------| -----| --------|
| OpenBLAS | 7.82 | 8.62 | 10.34 |
| MKLML | 11.02 | 12.86 | 15.33 |
| MKL-DNN | 27.69 | 28.8 | 29.27 |
chart on batch size 128
TBD
- ResNet
- GoogLeNet
### Laptop
TBD
### Desktop
TBD
benchmark/paddle/image/resnet.py 0 → 100644
浏览文件 @ 06c7686a
#!/usr/bin/env python
from paddle.trainer_config_helpers import *
height = 224
width = 224
num_class = 1000
batch_size = get_config_arg('batch_size', int, 64)
layer_num = get_config_arg("layer_num", int, 50)
is_test = get_config_arg("is_test", bool, False)
args = {'height': height, 'width': width, 'color': True, 'num_class': num_class}
define_py_data_sources2(
"train.list", None, module="provider", obj="process", args=args)
settings(
batch_size=batch_size,
learning_rate=0.01 / batch_size,
learning_method=MomentumOptimizer(0.9),
regularization=L2Regularization(0.0005 * batch_size))
#######################Network Configuration #############
def conv_bn_layer(name,
input,
filter_size,
num_filters,
stride,
padding,
channels=None,
active_type=ReluActivation()):
"""
A wrapper for conv layer with batch normalization layers.
Note:
conv layer has no activation.
"""
tmp = img_conv_layer(
name=name + "_conv",
input=input,
filter_size=filter_size,
num_channels=channels,
num_filters=num_filters,
stride=stride,
padding=padding,
act=LinearActivation(),
bias_attr=False)
return batch_norm_layer(
name=name + "_bn", input=tmp, act=active_type, use_global_stats=is_test)
def bottleneck_block(name, input, num_filters1, num_filters2):
"""
A wrapper for bottlenect building block in ResNet.
Last conv_bn_layer has no activation.
Addto layer has activation of relu.
"""
last_name = conv_bn_layer(
name=name + '_branch2a',
input=input,
filter_size=1,
num_filters=num_filters1,
stride=1,
padding=0)
last_name = conv_bn_layer(
name=name + '_branch2b',
input=last_name,
filter_size=3,
num_filters=num_filters1,
stride=1,
padding=1)
last_name = conv_bn_layer(
name=name + '_branch2c',
input=last_name,
filter_size=1,
num_filters=num_filters2,
stride=1,
padding=0,
active_type=LinearActivation())
return addto_layer(
name=name + "_addto", input=[input, last_name], act=ReluActivation())
def mid_projection(name, input, num_filters1, num_filters2, stride=2):
"""
A wrapper for middile projection in ResNet.
projection shortcuts are used for increasing dimensions,
and other shortcuts are identity
branch1: projection shortcuts are used for increasing
dimensions, has no activation.
branch2x: bottleneck building block, shortcuts are identity.
"""
# stride = 2
branch1 = conv_bn_layer(
name=name + '_branch1',
input=input,
filter_size=1,
num_filters=num_filters2,
stride=stride,
padding=0,
active_type=LinearActivation())
last_name = conv_bn_layer(
name=name + '_branch2a',
input=input,
filter_size=1,
num_filters=num_filters1,
stride=stride,
padding=0)
last_name = conv_bn_layer(
name=name + '_branch2b',
input=last_name,
filter_size=3,
num_filters=num_filters1,
stride=1,
padding=1)
last_name = conv_bn_layer(
name=name + '_branch2c',
input=last_name,
filter_size=1,
num_filters=num_filters2,
stride=1,
padding=0,
active_type=LinearActivation())
return addto_layer(
name=name + "_addto", input=[branch1, last_name], act=ReluActivation())
img = data_layer(name='image', size=height * width * 3)
def deep_res_net(res2_num=3, res3_num=4, res4_num=6, res5_num=3):
"""
A wrapper for 50,101,152 layers of ResNet.
res2_num: number of blocks stacked in conv2_x
res3_num: number of blocks stacked in conv3_x
res4_num: number of blocks stacked in conv4_x
res5_num: number of blocks stacked in conv5_x
"""
# For ImageNet
# conv1: 112x112
tmp = conv_bn_layer(
"conv1",
input=img,
filter_size=7,
channels=3,
num_filters=64,
stride=2,
padding=3)
tmp = img_pool_layer(name="pool1", input=tmp, pool_size=3, stride=2)
# conv2_x: 56x56
tmp = mid_projection(
name="res2_1", input=tmp, num_filters1=64, num_filters2=256, stride=1)
for i in xrange(2, res2_num + 1, 1):
tmp = bottleneck_block(
name="res2_" + str(i), input=tmp, num_filters1=64, num_filters2=256)
# conv3_x: 28x28
tmp = mid_projection(
name="res3_1", input=tmp, num_filters1=128, num_filters2=512)
for i in xrange(2, res3_num + 1, 1):
tmp = bottleneck_block(
name="res3_" + str(i),
input=tmp,
num_filters1=128,
num_filters2=512)
# conv4_x: 14x14
tmp = mid_projection(
name="res4_1", input=tmp, num_filters1=256, num_filters2=1024)
for i in xrange(2, res4_num + 1, 1):
tmp = bottleneck_block(
name="res4_" + str(i),
input=tmp,
num_filters1=256,
num_filters2=1024)
# conv5_x: 7x7
tmp = mid_projection(
name="res5_1", input=tmp, num_filters1=512, num_filters2=2048)
for i in xrange(2, res5_num + 1, 1):
tmp = bottleneck_block(
name="res5_" + str(i),
input=tmp,
num_filters1=512,
num_filters2=2048)
tmp = img_pool_layer(
name='avgpool',
input=tmp,
pool_size=7,
stride=1,
pool_type=AvgPooling())
return fc_layer(input=tmp, size=num_class, act=SoftmaxActivation())
if layer_num == 50:
resnet = deep_res_net(3, 4, 6, 3)
elif layer_num == 101:
resnet = deep_res_net(3, 4, 23, 3)
elif layer_num == 152:
resnet = deep_res_net(3, 8, 36, 3)
else:
print("Wrong layer number.")
lbl = data_layer(name="label", size=num_class)
loss = cross_entropy(name='loss', input=resnet, label=lbl)
inputs(img, lbl)
outputs(loss)
benchmark/paddle/image/run_mkldnn.sh
浏览文件 @ 06c7686a
......@@ -5,22 +5,23 @@ function train() {
export OMP_DYNAMIC="FALSE"
export KMP_AFFINITY="granularity=fine,compact,0,0"
topology=$1
bs=$2
use_mkldnn=$3
if [ $3 == "True" ]; then
layer_num=$2
bs=$3
use_mkldnn=$4
if [ $4 == "True" ]; then
thread=1
log="logs/${topology}-mkldnn-${bs}.log"
elif [ $3 == "False" ]; then
log="logs/${topology}-${layer_num}-mkldnn-${bs}.log"
elif [ $4 == "False" ]; then
thread=`nproc`
# each trainer_count use only 1 core to avoid conflict
export OMP_NUM_THREADS=1
export MKL_NUM_THREADS=1
log="logs/${topology}-${thread}mklml-${bs}.log"
log="logs/${topology}-${layer_num}-${thread}mklml-${bs}.log"
else
echo "Wrong input $3, use True or False."
exit 0
fi
args="batch_size=${bs}"
args="batch_size=${bs},layer_num=${layer_num}"
config="${topology}.py"
paddle train --job=time \
--config=$config \
......@@ -40,12 +41,9 @@ if [ ! -d "logs" ]; then
mkdir logs
fi
#========== mkldnn ==========#
train vgg 64 True
train vgg 128 True
train vgg 256 True
#========== mklml ===========#
train vgg 64 False
train vgg 128 False
train vgg 256 False
for use_mkldnn in True False; do
for batchsize in 64 128 256; do
train vgg 19 $batchsize $use_mkldnn
train resnet 50 $batchsize $use_mkldnn
done
done
benchmark/paddle/image/vgg.py
浏览文件 @ 06c7686a
......@@ -13,7 +13,7 @@ define_py_data_sources2(
settings(
batch_size=batch_size,
learning_rate=0.01 / batch_size,
learning_rate=0.001 / batch_size,
learning_method=MomentumOptimizer(0.9),
regularization=L2Regularization(0.0005 * batch_size))
......
cmake/cblas.cmake
浏览文件 @ 06c7686a
# Find the CBlas and lapack libraries
#
# It will search MKL, atlas, OpenBlas, reference-cblas in order.
# It will search MKLML, atlas, OpenBlas, reference-cblas in order.
#
# If any cblas implementation found, the following variable will be set.
# CBLAS_PROVIDER # one of MKL, ATLAS, OPENBLAS, REFERENCE
# CBLAS_PROVIDER # one of MKLML, ATLAS, OPENBLAS, REFERENCE
# CBLAS_INC_DIR # the include directory for cblas.
# CBLAS_LIBS # a list of libraries should be linked by paddle.
# # Each library should be full path to object file.
#
# User should set one of MKL_ROOT, ATLAS_ROOT, OPENBLAS_ROOT, REFERENCE_CBLAS_ROOT
# during cmake. If none of them set, it will try to find cblas implementation in
# system paths.
#
set(CBLAS_FOUND OFF)
......@@ -30,44 +25,6 @@ if(WITH_MKLML AND MKLML_INC_DIR AND MKLML_LIB)
return()
endif()
## Then find MKL.
set(INTEL_MKL_ROOT "/opt/intel/mkl" CACHE PATH "Folder contains intel mkl libs")
set(MKL_ROOT $ENV{MKL_ROOT} CACHE PATH "Folder contains env MKL")
set(MKL_INCLUDE_SEARCH_PATHS
${MKL_ROOT}/include
${INTEL_MKL_ROOT}/include)
set(MKL_LIB_SEARCH_PATHS
${MKL_ROOT}/lib
${MKL_ROOT}/lib/intel64
${INTEL_MKL_ROOT}/lib
${INTEL_MKL_ROOT}/lib/intel64)
find_path(MKL_INC_DIR mkl.h PATHS
${MKL_INCLUDE_SEARCH_PATHS})
find_path(MKL_LAPACK_INC_DIR mkl_lapacke.h PATHS
${MKL_INCLUDE_SEARCH_PATHS})
find_library(MKL_CORE_LIB NAMES mkl_core PATHS
${MKL_LIB_SEARCH_PATHS})
find_library(MKL_SEQUENTIAL_LIB NAMES mkl_sequential PATHS
${MKL_LIB_SEARCH_PATHS})
find_library(MKL_INTEL_LP64 NAMES mkl_intel_lp64 PATHS
${MKL_LIB_SEARCH_PATHS})
if(MKL_LAPACK_INC_DIR AND MKL_INC_DIR AND MKL_CORE_LIB AND MKL_SEQUENTIAL_LIB AND MKL_INTEL_LP64)
set(CBLAS_FOUND ON)
set(CBLAS_PROVIDER MKL)
set(CBLAS_INC_DIR ${MKL_INC_DIR} ${MKL_LAPACK_INC_DIR})
set(CBLAS_LIBRARIES ${MKL_INTEL_LP64} ${MKL_SEQUENTIAL_LIB} ${MKL_CORE_LIB})
add_definitions(-DPADDLE_USE_MKL)
add_definitions(-DLAPACK_FOUND)
message(STATUS "Found MKL (include: ${MKL_INC_DIR}, library: ${CBLAS_LIBRARIES})")
message(STATUS "Found lapack in MKL (include: ${MKL_LAPACK_INC_DIR})")
return()
endif()
## Then find atlas.
set(ATLAS_ROOT $ENV{ATLAS_ROOT} CACHE PATH "Folder contains Atlas")
set(ATLAS_INCLUDE_SEARCH_PATHS
......
cmake/configure.cmake
浏览文件 @ 06c7686a
......@@ -62,11 +62,11 @@ else()
FIND_PACKAGE(CUDA REQUIRED)
if(${CUDA_VERSION_MAJOR} VERSION_LESS 7)
message(FATAL_ERROR "Paddle need CUDA >= 7.0 to compile")
message(FATAL_ERROR "Paddle needs CUDA >= 7.0 to compile")
endif()
if(NOT CUDNN_FOUND)
message(FATAL_ERROR "Paddle need cudnn to compile")
message(FATAL_ERROR "Paddle needs cudnn to compile")
endif()
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${SIMD_FLAG}")
......
cmake/cross_compiling/ios.cmake
浏览文件 @ 06c7686a
......@@ -79,9 +79,8 @@ if(NOT DEFINED IOS_ARCH)
# FIXME(liuyiqun): support "armv7;armv7s;arm64" future
set(IOS_ARCH "arm64")
elseif(IOS_PLATFORM STREQUAL "SIMULATOR")
set(IOS_ARCH "i386;x86_64")
elseif(IOS_PLATFORM STREQUAL "WATCHOS")
set(IOS_ARCH armv7k)
# FIXME(liuyiqun): support "i386;x86_64" future
set(IOS_ARCH "x86_64")
endif()
endif()
set(CMAKE_OSX_ARCHITECTURES ${IOS_ARCH} CACHE string "Build architecture for iOS")
......
cmake/external/eigen.cmake
浏览文件 @ 06c7686a
......@@ -8,7 +8,7 @@ ExternalProject_Add(
extern_eigen3
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/RLovelett/eigen.git"
GIT_TAG 4e79cb69b9425f5f8c3a84be4350d4ab75b5fd9d
GIT_TAG 70661066beef694cadf6c304d0d07e0758825c10
PREFIX ${EIGEN_SOURCE_DIR}
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
......
cmake/external/mkldnn.cmake
浏览文件 @ 06c7686a
......@@ -46,16 +46,20 @@ IF(${CBLAS_PROVIDER} STREQUAL "MKLML")
MESSAGE(STATUS "Build MKLDNN with ${MKLDNN_MKLROOT}")
ENDIF()
SET(MKLDNN_CFLAG "${CMAKE_C_FLAGS} -Wno-error=strict-overflow")
SET(MKLDNN_CXXFLAG "${CMAKE_CXX_FLAGS} -Wno-error=strict-overflow")
ExternalProject_Add(
${MKLDNN_PROJECT}
${EXTERNAL_PROJECT_LOG_ARGS}
DEPENDS ${MKLDNN_DEPENDS}
GIT_REPOSITORY "https://github.com/01org/mkl-dnn.git"
GIT_TAG "v0.10"
GIT_TAG "v0.11"
PREFIX ${MKLDNN_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL_DIR}
CMAKE_ARGS -DMKLROOT=${MKLDNN_MKLROOT}
CMAKE_ARGS -DCMAKE_C_FLAGS=${MKLDNN_CFLAG}
CMAKE_ARGS -DCMAKE_CXX_FLAGS=${MKLDNN_CXXFLAG}
CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${MKLDNN_INSTALL_DIR}
-DMKLROOT:PATH=${MKLDNN_MKLROOT}
)
......
cmake/external/mklml.cmake
浏览文件 @ 06c7686a
......@@ -27,8 +27,8 @@ ENDIF()
INCLUDE(ExternalProject)
SET(MKLML_PROJECT "extern_mklml")
SET(MKLML_VER "mklml_lnx_2018.0.20170720")
SET(MKLML_URL "https://github.com/01org/mkl-dnn/releases/download/v0.10/${MKLML_VER}.tgz")
SET(MKLML_VER "mklml_lnx_2018.0.1.20171007")
SET(MKLML_URL "https://github.com/01org/mkl-dnn/releases/download/v0.11/${MKLML_VER}.tgz")
SET(MKLML_SOURCE_DIR "${THIRD_PARTY_PATH}/mklml")
SET(MKLML_DOWNLOAD_DIR "${MKLML_SOURCE_DIR}/src/${MKLML_PROJECT}")
SET(MKLML_DST_DIR "mklml")
......
cmake/external/nccl.cmake 0 → 100644
浏览文件 @ 06c7686a
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# 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.
if(NOT WITH_GPU)
return()
endif()
include(ExternalProject)
set(NCCL_SOURCE_DIR ${THIRD_PARTY_PATH}/nccl)
include_directories(${NCCL_SOURCE_DIR}/src/extern_nccl/src)
if(WITH_DSO)
# If we use DSO, we do not build nccl, just download the dependencies
set(NCCL_BUILD_COMMAND "")
set(NCCL_INSTALL_COMMAND "")
set(NCCL_INSTALL_DIR "")
else()
# otherwise, we build nccl and link it.
set(NCCL_INSTALL_DIR ${THIRD_PARTY_PATH}/install/nccl)
# Note: cuda 8.0 is needed to make nccl
# When cuda is not installed on the system directory, need to set CUDA_HOME to your cuda root
set(NCCL_BUILD_COMMAND "make -j 8")
set(NCCL_INSTALL_COMMAND "make install PREFIX=${NCCL_INSTALL_DIR}")
endif()
ExternalProject_Add(
extern_nccl
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/NVIDIA/nccl.git"
GIT_TAG "v1.3.4-1"
PREFIX "${NCCL_SOURCE_DIR}"
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_COMMAND "${NCCL_BUILD_COMMAND}"
INSTALL_COMMAND "${NCCL_INSTALL_COMMAND}"
INSTALL_DIR "${NCCL_INSTALL_DIR}"
TEST_COMMAND ""
)
if(WITH_DSO)
if(${CMAKE_VERSION} VERSION_LESS "3.3.0")
set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/lib_nccl_dummy.c)
file(WRITE ${dummyfile} "const char * dummy_nccl = \"${dummyfile}\";")
add_library(nccl STATIC ${dummyfile})
else()
add_library(nccl INTERFACE)
endif()
else()
add_library(nccl STATIC IMPORTED GLOBAL)
set_property(TARGET nccl PROPERTY IMPORTED_LOCATION
${NCCL_INSTALL_DIR}/lib/libnccl_static.a)
endif()
add_dependencies(nccl extern_nccl)
cmake/external/openblas.cmake
浏览文件 @ 06c7686a
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
#
# 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.
......@@ -86,7 +86,7 @@ IF(NOT ${CBLAS_FOUND})
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
)
SET(CBLAS_PROVIDER openblas)
IF(WITH_C_API)
INSTALL(DIRECTORY ${CBLAS_INC_DIR} DESTINATION third_party/openblas)
# Because libopenblas.a is a symbolic link of another library, thus need to
......@@ -115,7 +115,7 @@ INCLUDE_DIRECTORIES(${CBLAS_INC_DIR})
# linear algebra libraries for cc_library(xxx SRCS xxx.c DEPS cblas)
SET(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/cblas_dummy.c)
FILE(WRITE ${dummyfile} "const char * dummy = \"${dummyfile}\";")
IF(${CBLAS_PROVIDER} MATCHES MKL)
IF("${CBLAS_PROVIDER}" STREQUAL "MKLML")
ADD_LIBRARY(cblas SHARED ${dummyfile})
ELSE()
ADD_LIBRARY(cblas STATIC ${dummyfile})
......
cmake/external/pybind11.cmake
浏览文件 @ 06c7686a
INCLUDE(ExternalProject)
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# 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.
SET(PYBIND_SOURCE_DIR ${THIRD_PARTY_PATH}/pybind)
if(NOT WITH_PYTHON)
return()
endif()
include(ExternalProject)
INCLUDE_DIRECTORIES(${PYBIND_SOURCE_DIR}/src/extern_pybind/include)
set(PYBIND_SOURCE_DIR ${THIRD_PARTY_PATH}/pybind)
include_directories(${PYBIND_SOURCE_DIR}/src/extern_pybind/include)
ExternalProject_Add(
extern_pybind
......@@ -17,14 +35,12 @@ ExternalProject_Add(
TEST_COMMAND ""
)
if (${CMAKE_VERSION} VERSION_LESS "3.3.0")
if(${CMAKE_VERSION} VERSION_LESS "3.3.0")
set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/pybind_dummy.c)
file(WRITE ${dummyfile} "const char * dummy_any = \"${dummyfile}\";")
file(WRITE ${dummyfile} "const char * dummy_pybind = \"${dummyfile}\";")
add_library(pybind STATIC ${dummyfile})
else()
add_library(pybind INTERFACE)
endif()
add_dependencies(pybind extern_pybind)
LIST(APPEND external_project_dependencies pybind)
cmake/external/swig.cmake
浏览文件 @ 06c7686a
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
......
cmake/external/zlib.cmake
浏览文件 @ 06c7686a
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
......
cmake/generic.cmake
浏览文件 @ 06c7686a
......@@ -93,7 +93,7 @@ include_directories(${CMAKE_CURRENT_BINARY_DIR})
if(NOT APPLE AND NOT ANDROID)
find_package(Threads REQUIRED)
link_libraries(${CMAKE_THREAD_LIBS_INIT})
set(CMAKE_CXX_LINK_EXECUTABLE "${CMAKE_CXX_LINK_EXECUTABLE} -ldl -lrt")
set(CMAKE_CXX_LINK_EXECUTABLE "${CMAKE_CXX_LINK_EXECUTABLE} -pthread -ldl -lrt")
endif(NOT APPLE AND NOT ANDROID)
function(merge_static_libs TARGET_NAME)
......
cmake/simd.cmake
浏览文件 @ 06c7686a
# This file is use to check all support level of AVX on your machine
# so that PaddlePaddle can unleash the vectorization power of muticore.
INCLUDE(CheckCXXSourceRuns)
INCLUDE(CheckCXXSourceCompiles)
include(CheckCXXSourceRuns)
include(CheckCXXSourceCompiles)
IF(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
if(CMAKE_COMPILER_IS_GNUCC OR CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
set(MMX_FLAG "-mmmx")
set(SSE2_FLAG "-msse2")
set(SSE3_FLAG "-msse3")
SET(AVX_FLAG "-mavx")
SET(AVX2_FLAG "-mavx2")
ELSEIF(MSVC)
set(AVX_FLAG "-mavx")
set(AVX2_FLAG "-mavx2")
elseif(MSVC)
set(MMX_FLAG "/arch:MMX")
set(SSE2_FLAG "/arch:SSE2")
set(SSE3_FLAG "/arch:SSE3")
SET(AVX_FLAG "/arch:AVX")
SET(AVX2_FLAG "/arch:AVX2")
ENDIF()
endif()
set(CMAKE_REQUIRED_FLAGS_RETAINED ${CMAKE_REQUIRED_FLAGS})
# Check MMX
set(CMAKE_REQUIRED_FLAGS ${MMX_FLAG})
set(MMX_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE)
CHECK_CXX_SOURCE_RUNS("
#include <mmintrin.h>
int main()
......@@ -32,6 +33,7 @@ int main()
# Check SSE2
set(CMAKE_REQUIRED_FLAGS ${SSE2_FLAG})
set(SSE2_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE)
CHECK_CXX_SOURCE_RUNS("
#include <emmintrin.h>
int main()
......@@ -42,6 +44,7 @@ int main()
# Check SSE3
set(CMAKE_REQUIRED_FLAGS ${SSE3_FLAG})
set(SSE3_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE)
CHECK_CXX_SOURCE_RUNS("
#include <pmmintrin.h>
int main()
......@@ -55,6 +58,7 @@ int main()
# Check AVX
set(CMAKE_REQUIRED_FLAGS ${AVX_FLAG})
set(AVX_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE)
CHECK_CXX_SOURCE_RUNS("
#include <immintrin.h>
int main()
......@@ -67,6 +71,7 @@ int main()
# Check AVX 2
set(CMAKE_REQUIRED_FLAGS ${AVX2_FLAG})
set(AVX2_FOUND_EXITCODE 1 CACHE STRING "Result from TRY_RUN" FORCE)
CHECK_CXX_SOURCE_RUNS("
#include <immintrin.h>
int main()
......
doc/api/v2/config/layer.rst
浏览文件 @ 06c7686a
......@@ -82,6 +82,11 @@ maxout
.. autoclass:: paddle.v2.layer.maxout
:noindex:
roi_pool
--------
.. autoclass:: paddle.v2.layer.roi_pool
:noindex:
Norm Layer
==========
......
doc/api/v2/data.rst
浏览文件 @ 06c7686a
......@@ -2,112 +2,9 @@
Data Reader Interface and DataSets
==================================
.. toctree::
:maxdepth: 1
DataTypes
=========
.. automodule:: paddle.v2.data_type
:members:
:noindex:
DataFeeder
==========
.. automodule:: paddle.v2.data_feeder
:members:
:noindex:
Reader
======
.. automodule:: paddle.v2.reader
:members:
:noindex:
.. automodule:: paddle.v2.reader.creator
:members:
:noindex:
minibatch
=========
.. automodule:: paddle.v2.minibatch
:members:
:noindex:
Dataset
=======
.. automodule:: paddle.v2.dataset
:members:
:noindex:
mnist
+++++
.. automodule:: paddle.v2.dataset.mnist
:members:
:noindex:
cifar
+++++
.. automodule:: paddle.v2.dataset.cifar
:members:
:noindex:
conll05
+++++++
.. automodule:: paddle.v2.dataset.conll05
:members: get_dict,get_embedding,test
:noindex:
imdb
++++
.. automodule:: paddle.v2.dataset.imdb
:members:
:noindex:
imikolov
++++++++
.. automodule:: paddle.v2.dataset.imikolov
:members:
:noindex:
movielens
+++++++++
.. automodule:: paddle.v2.dataset.movielens
:members:
:noindex:
.. autoclass:: paddle.v2.dataset.movielens.MovieInfo
:noindex:
.. autoclass:: paddle.v2.dataset.movielens.UserInfo
:noindex:
sentiment
+++++++++
.. automodule:: paddle.v2.dataset.sentiment
:members:
:noindex:
uci_housing
+++++++++++
.. automodule:: paddle.v2.dataset.uci_housing
:members:
:noindex:
wmt14
+++++
.. automodule:: paddle.v2.dataset.wmt14
:members:
:noindex:
data/data_reader.rst
data/image.rst
data/dataset.rst
doc/api/v2/data/data_reader.rst 0 → 100644
浏览文件 @ 06c7686a
=====================
Data Reader Interface
=====================
DataTypes
=========
.. automodule:: paddle.v2.data_type
:members:
:noindex:
DataFeeder
==========
.. automodule:: paddle.v2.data_feeder
:members:
:noindex:
Reader
======
.. automodule:: paddle.v2.reader
:members:
:noindex:
.. automodule:: paddle.v2.reader.creator
:members:
:noindex:
minibatch
=========
.. automodule:: paddle.v2.minibatch
:members:
:noindex:
doc/api/v2/data/dataset.rst 0 → 100644
浏览文件 @ 06c7686a
Dataset
=======
.. automodule:: paddle.v2.dataset
:members:
:noindex:
mnist
+++++
.. automodule:: paddle.v2.dataset.mnist
:members:
:noindex:
cifar
+++++
.. automodule:: paddle.v2.dataset.cifar
:members:
:noindex:
conll05
+++++++
.. automodule:: paddle.v2.dataset.conll05
:members: get_dict,get_embedding,test
:noindex:
imdb
++++
.. automodule:: paddle.v2.dataset.imdb
:members:
:noindex:
imikolov
++++++++
.. automodule:: paddle.v2.dataset.imikolov
:members:
:noindex:
movielens
+++++++++
.. automodule:: paddle.v2.dataset.movielens
:members:
:noindex:
.. autoclass:: paddle.v2.dataset.movielens.MovieInfo
:noindex:
.. autoclass:: paddle.v2.dataset.movielens.UserInfo
:noindex:
sentiment
+++++++++
.. automodule:: paddle.v2.dataset.sentiment
:members:
:noindex:
uci_housing
+++++++++++
.. automodule:: paddle.v2.dataset.uci_housing
:members:
:noindex:
wmt14
+++++
.. automodule:: paddle.v2.dataset.wmt14
:members:
:noindex:
doc/api/v2/data/image.rst 0 → 100644
浏览文件 @ 06c7686a
Image Interface
===============
.. automodule:: paddle.v2.image
:members:
doc/design/block.md
浏览文件 @ 06c7686a
......@@ -189,7 +189,7 @@ OpDesc {
inputs = {0} // the index of x in vars of BlockDesc above
outputs = {5, 3} // indices of act and hidden_out in vars of BlockDesc above
attrs {
"memories" : {1} // the index of h
"states" : {1} // the index of h
"step_net" : <above step net>
}
};
......
doc/design/evaluator.md 0 → 100644
浏览文件 @ 06c7686a
## Evaluator Design
### The Problem
During training or serving, we provide the evaluation function to measure the model performance, e.g., accuracy, precision. In the operator based framework design, the data go through the network pipeline batch by batch. As a result, inside the operator, we only can calculate one minibatch metrics. We need to provide a mechanism to calculate the metrics for each N pass/batch the user wanted.
### Evaluator Design
Currently, every operation is expressed in the graph. we divide the evaluator process into three steps.
1. Initialize the metric state and add it into the block.
2. Calculate the statistic of the metric state in every mini-batch. The single operator is only responsible for calculating necessary statistics for one mini-batch. For example, accuracy operator only calculate a minibatch data if run once.
3. Merge the mini-batch statistics to form the evaluation result for multiple mini-batches. When it comes to distributed training/Multi-GPU training, aggregate the value from different devices.
### Implementation
This design is shown in python API.
Each metric operator need to caculate the metric statistic and return the batch aware states, Python side responsible for accumulate the states for each pass.
```python
class Evaluator(object):
"""
Evaluator Base class.
"""
def __init__(self, name, **kwargs):
"""
Different evaluator may has different metric states. E.g, Accuracy need two variables, total and right sample counts.
Auc need four variables, `true_positives`,
`true_negatives`, `false_positives` and `false_negatives`. So every evaluator should create its needed variables and append to main_program
The initialization of Evaluator should be responsible for:
create metric states and append to the main_program
"""
pass
def _update_ops(self, input, label, **kwargs)
"""
Add mini-batch evaluator caculate operators to the main_program.
Add increment operator to accumulate the metric states.
"""
def reset(self, executor, reset_program=None):
"""
Reset metric states at the begin of each pass/user specified batch number.
Execute the reset_program to reset the states.
"""
def eval(self, executor, eval_program=None):
"""
Merge the mini-batch statistics to form the evaluation result for multiple mini-batches.
Execute the eval_program and return the result.
"""
return eval_result
```
doc/design/float16.md 0 → 100644
浏览文件 @ 06c7686a
# Design Doc: float16
## Why float16
Half precision (float16) is a binary floating-point format that occupies 16 bits in memory. float16 is half the size of traditional 32-bit single precision format (float) and has lower precision and smaller range.
When high precision computation is not required, using float16 data type could potentially
- reduce storage space, memory bandwidth, and power usages;
- increase the chance of data fitting into a smaller cache of lower latency;
- provide arithmetic speed up if supported by hardware.
## Survey of current float16 support
A brief survey of float16 support on different compilers, hardwares, and libraries can be found below. Interested readers can refer to [link1](https://github.com/PaddlePaddle/Paddle/issues/4853) and [link2](https://github.com/Xreki/Xreki.github.io/blob/master/multi_data_types_in_dl_framework/ppt/float16_and_quantized_type.md) for more info.
The goal of float16 is to serve as a key for the executor to find and run the correct version of compute method specialized for float16 in operator kernel. It should be compatible with various natively supported float16 implementations including `__half` for cuda, `float16_t` for ARM, and `Eigen::half` for Eigen to make writing customized float16 kernels easier.
### Compiler
- nvcc supports `__half` data type after CUDA 7.5.
- `__fp16` or `float16_t` is supported as storage type for gcc >= 6.1 and clang >= 3.4.
- `__fp16` or `float16_t` is supported as arithmetic type for gcc >= 7.1 and clang >= 3.9.
### Hardware
- `__half` is supported on GPU with compute capability >= 5.3.
- `__fp16` is supported as storage type for ARMv7-A, ARMv8-A, and above.
- `__fp16` is supported as arithmetic type after ARMv8.2-A (currently, the only microarchitecture implementing ARMv8.2-A is ARM Cortex-A75, which is announced in May 2017. There seems to be no application processors currently available on market that adopts this architecture. It is reported that Qualcomm Snapdragon 845 uses Cortex-A75 design and will be available in mobile devices in early 2018).
### Libraries
- [Eigen](https://github.com/RLovelett/eigen) >= 3.3 supports float16 calculation on both GPU and CPU using the `Eigen::half` class. It is mostly useful for Nvidia GPUs because of the overloaded arithmetic operators using cuda intrinsics. It falls back to using software emulation on CPU for calculation and there is no special treatment to ARM processors.
- [ARM compute library](https://github.com/ARM-software/ComputeLibrary) >= 17.02.01 supports NEON FP16 kernels (requires ARMv8.2-A CPU).
## Implementation
The float16 class holds a 16-bit `uint16_t` data internally.
```
struct float16 {
uint16_t x;
};
```
float16 supports the following features:
- constructors / assignment operators that take input from primitive data types including bool, integers of various length, float, and double.
- constructors / assignment operators that take input from `__half` on cuda, `float16_t` on ARM, and `Eigen::half` on Eigen.
- conversion operators to primitive data types and half precision data types on cuda, ARM and Eigen.
- overloaded arithmetic operators for cuda, arm, and non-arm cpu, respectively. These operators will take advantage of the cuda and ARM intrinsics on the corresponding hardware.
To support the above features, two fundamental conversion functions are provided:
```
float16 float_to_half_rn(float f); // convert to half precision in round-to-nearest-even mode
float half_to_float(float16 h);
```
which provides one-to-one conversion between float32 and float16. These twos functions will do different conversion routines based on the current hardware. CUDA/ARM instrinsics will be used when the corresonding hardware is available. If the hardware or compiler level does not support float32 to float16 conversion, software emulation will be performed to do the conversion.
## To do
After float16 class is available, some of the future items are below:
- Update pybind/tensor_py.h to bind c++ float16 with numpy float16.
- Modify `GetKernelType()` method in `framework/operator.h` to make it compatible with float16.
- Create a type-casting operator that can convert the data type in tensor between float16 and other types.
doc/design/graph_survey.md 0 → 100644
浏览文件 @ 06c7686a
## Survey on Graph
Neural network framework often provides symbolic API for users to write network topology conveniently. This doc manily focus on symbolic API in most popular neural network frameworks, and try to find out how to parse symbolic configuration to a portable file, such as protobuf or json.
### Mxnet
The core concept of symbolic API is `Symbol`. Mxnet implements `Symbol` class in C++, and export to Python using C-API. Please refer to the comments in Mxnet:
`Symbol` is help class used to represent the operator node in Graph.
`Symbol` acts as an interface for building graphs from different components like Variable, Functor and Group. `Symbol` is also exported to python front-end (while Graph is not) to enable quick test and deployment. Conceptually, symbol is the final operation of a graph and thus including all the information required (the graph) to evaluate its output value.
A simple network topology wrote by Symbol is as follows:
```python
def get_symbol(num_classes=10, **kwargs):
data = mx.symbol.Variable('data')
data = mx.symbol.Flatten(data=data)
fc1 = mx.symbol.FullyConnected(data = data, name='fc1', num_hidden=128)
act1 = mx.symbol.Activation(data = fc1, name='relu1', act_type="relu")
fc2 = mx.symbol.FullyConnected(data = act1, name = 'fc2', num_hidden = 64)
act2 = mx.symbol.Activation(data = fc2, name='relu2', act_type="relu")
fc3 = mx.symbol.FullyConnected(data = act2, name='fc3', num_hidden=num_classes)
mlp = mx.symbol.SoftmaxOutput(data = fc3, name = 'softmax')
return mlp
```
Varible here is actually a Symbol. Every basic Symbol will correspond to one Node, and every Node has its own NodeAttr. There is a op field in NodeAttr class, when a Symbol represents Variable(often input data), the op field is null.
Symbol contains a data member, std::vector<NodeEntry> outputs, and NodeEntry cantains a poniter to Node. We can follow the Node pointer to get all the Graph.
And Symbol can be saved to a Json file.
Here is a detailed example:
```
>>> import mxnet as mx
>>> data = mx.symbol.Variable('data')
>>> print data.debug_str()
Variable:data
>>> data = mx.symbol.Flatten(data=data)
>>> print data.debug_str()
Symbol Outputs:
output[0]=flatten0(0)
Variable:data
--------------------
Op:Flatten, Name=flatten0
Inputs:
arg[0]=data(0) version=0
>>> fc1 = mx.symbol.FullyConnected(data = data, name='fc1', num_hidden=128)
>>> print fc1.debug_str()
Symbol Outputs:
output[0]=fc1(0)
Variable:data
--------------------
Op:Flatten, Name=flatten0
Inputs:
arg[0]=data(0) version=0
Variable:fc1_weight
Variable:fc1_bias
--------------------
Op:FullyConnected, Name=fc1
Inputs:
arg[0]=flatten0(0)
arg[1]=fc1_weight(0) version=0
arg[2]=fc1_bias(0) version=0
Attrs:
num_hidden=128
```
### TensorFlow
The core concept of symbolic API is `Tensor`. Tensorflow defines `Tensor` in Python. Please refer to the comments in TensorFlow:
A `Tensor` is a symbolic handle to one of the outputs of an `Operation`. It does not hold the values of that operation's output, but instead provides a means of computing those values in a TensorFlow [Session](https://www.tensorflow.org/api_docs/python/tf/Session).
A simple example is as follows:
```python
# Build a dataflow graph.
c = tf.constant([[1.0, 2.0], [3.0, 4.0]])
d = tf.constant([[1.0, 1.0], [0.0, 1.0]])
e = tf.matmul(c, d)
# Construct a `Session` to execute the graph.
sess = tf.Session()
# Execute the graph and store the value that `e` represents in `result`.
result = sess.run(e)
```
The main method of `Tensor` is as follows:
```python
@property
def op(self):
"""The `Operation` that produces this tensor as an output."""
return self._op
@property
def dtype(self):
"""The `DType` of elements in this tensor."""
return self._dtype
@property
def graph(self):
"""The `Graph` that contains this tensor."""
return self._op.graph
@property
def name(self):
"""The string name of this tensor."""
if not self._op.name:
raise ValueError("Operation was not named: %s" % self._op)
return "%s:%d" % (self._op.name, self._value_index)
@property
def device(self):
"""The name of the device on which this tensor will be produced, or None."""
return self._op.device
```
Tensor can be taken as target to run by session. Tensor contains all the information of Graph, and tracks data dependency.
Here is a detailed example:
```
>>> import tensorflow as tf
>>> c = tf.constant([[1.0, 2.0], [3.0, 4.0]])
>>> print c.graph
<tensorflow.python.framework.ops.Graph object at 0x10f256d50>
>>> d = tf.constant([[1.0, 1.0], [0.0, 1.0]])
>>> print d.graph
<tensorflow.python.framework.ops.Graph object at 0x10f256d50>
>>> e = tf.matmul(c, d)
>>> print e.graph
<tensorflow.python.framework.ops.Graph object at 0x10f256d50>
```
### Dynet
The core concept of symbolic API is `Expression`, and Dynet defines `Expression` class in C++.
A simple example is as follows:
```cpp
ComputationGraph cg;
Expression W = parameter(cg, pW);
Expression in = input(cg, xs[i]);
Expression label = input(cg, ys[i]);
Expression pred = W * in;
Expression loss = square(pred - label);
```
The input data and parameter are also represented by Expression. Every basci Expression corresponds to a Node. And input data is also a Node.
Expression has a data member ComputationGraph, and ComputationGraph will be modified in users' configuring process. Expression can be a running target, beacuse Expression contains all dependency.
Here is a detailed example:
write topology in C++
```
ComputationGraph cg;
Expression W = parameter(cg, pW);
cg.print_graphviz();
Expression pred = W * xs[i];
cg.print_graphviz();
Expression loss = square(pred - ys[i]);
cg.print_graphviz();
```
compile and print
```
# first print
digraph G {
rankdir=LR;
nodesep=.05;
N0 [label="v0 = parameters({1}) @ 0x7ffe4de00110"];
}
# second print
digraph G {
rankdir=LR;
nodesep=.05;
N0 [label="v0 = parameters({1}) @ 0x7ffe4de00110"];
N1 [label="v1 = v0 * -0.98"];
N0 -> N1;
}
# third print
digraph G {
rankdir=LR;
nodesep=.05;
N0 [label="v0 = parameters({1}) @ 0x7ffe4de00110"];
N1 [label="v1 = v0 * -0.98"];
N0 -> N1;
N2 [label="v2 = -1.88387 - v1"];
N1 -> N2;
N3 [label="v3 = -v2"];
N2 -> N3;
N4 [label="v4 = square(v3)"];
N3 -> N4;
}
```
### Conclusion
Actually, Symbol/Tensor/Expression in Mxnet/TensorFlow/Dynet are the same level concepts. We use a unified name Expression here, this level concept has following features:
- Users wirte topoloy with symbolic API, and all return value is Expression, including input data and parameter.
- Expression corresponds with a global Graph, and Expression can also be composed.
- Expression tracks all dependency and can be taken as a run target
doc/design/mkldnn/README.MD
浏览文件 @ 06c7686a
......@@ -15,6 +15,7 @@
- [CMake](#cmake)
- [Layers](#layers)
- [Activations](#activations)
- [Weights](#weights)
- [Unit Tests](#unit-tests)
- [Protobuf Messages](#protobuf-messages)
- [Python API](#python-api)
......@@ -45,17 +46,23 @@ Figure 1. PaddlePaddle on IA.
### Layers
所有MKL-DNN相关的C++ layers，都会按照PaddlePaddle的目录结构存放在
`paddle/gserver/layers`中，并且文件名都会一以*Mkldnn*开头。
`paddle/gserver/layers`中，并且文件名都会一以*MKLDNN*开头。
所有MKL-DNN的layers都会继承于一个叫做`MkldnnLayer`的父类，该父类继承于PaddlePaddle的基类`Layer`。
所有MKL-DNN的layers都会继承于一个叫做`MKLDNNLayer`的父类，该父类继承于PaddlePaddle的基类`Layer`。
在`MKLDNNLayer`中会提供一些必要的接口和函数，并且会写好`forward`和`backward`的基本逻辑。部分函数定义为纯虚函数，子类只需要实现这些函数即可。
### Activations
由于在PaddlePaddle中，激活函数是独立于layer概念的，所以会在`paddle/gserver/activations`目录下添加一个`MkldnnActivation.h`文件定义一些用于MKL-DNN的接口，实现方法还是会在`ActivationFunction.cpp`文件。
由于在PaddlePaddle中，激活函数是独立于layer概念的，所以会在`paddle/gserver/activations`目录下添加`MKLDNNActivation.h`和`MKLDNNActivation.cpp`文件用于定义和使用MKL-DNN的接口。
### Unit Tests
会在`paddle/gserver/test`目录下添加`test_Mkldnn.cpp`和`MkldnnTester.*`用于MKL-DNN的测试。
### Weights
由于有些layer是含有参数的，我们会尽量让MKL-DNN的参数与PaddlePaddle中`parameter`共享一块内存。
同时，由于MKL-DNN在训练时使用的参数layout可能与PaddlePaddle默认的`nchw`不一致，我们会在网络训练的开始和结束时分别转换这个layout，使得最终保存的参数格式与PaddlePaddle一致。
Activation的测试，计划在PaddlePaddle原有的测试文件上直接添加新的测试type。
### Unit Tests
会在`paddle/gserver/test`目录下添加`test_MKLDNN.cpp`和`MKLDNNTester.*`用于MKL-DNN的测试。
测试分为每个layer(或activation)的单元测试和简单网络的整体测试。
每个测试会对比PaddlePaddle中CPU算出的结果与MKL-DNN的结果，小于某个比较小的阈值认为通过。
### Protobuf Messages
根据具体layer的需求可能会在`proto/ModelConfig.proto`里面添加必要的选项。
......@@ -82,7 +89,7 @@ if use_mkldnn
会在`v1_api_demo`目录下添加一个`mkldnn`的文件夹，里面放入一些用于MKL-DNN测试的demo脚本。
### Benchmarking
会考虑添加部分逻辑在`benchmark/paddle/image/run.sh`，添加使用MKL-DNN的测试。
会添加`benchmark/paddle/image/run_mkldnn.sh`，用于测试使用MKL-DNN之后的性能。
### Others
1. 如果在使用MKL-DNN的情况下，会把CPU的Buffer对齐为64。
......@@ -94,14 +101,16 @@ if use_mkldnn
我们总结出一些特别需要注意的点：
1. 使用**deviceId_**。为了尽可能少的在父类Layer中添加变量或者函数，我们决定使用已有的`deviceId_`变量来区分layer的属性，定义`-2`为`MkldnnLayer`特有的设备ID。
1. 使用**deviceId_**。为了尽可能少的在父类Layer中添加变量或者函数，我们决定使用已有的`deviceId_`变量来区分layer的属性，定义`-2`为`MKLDNNLayer`特有的设备ID。
2. 重写父类Layer的**init**函数，修改`deviceId_`为`-2`，代表这个layer是用于跑在MKL-DNN的环境下。
3. 创建`MkldnnMatrix`，用于管理MKL-DNN会用到的相关memory函数、接口以及会用的到格式信息。
4. 创建`MkldnnBase`，定义一些除了layer和memory相关的类和函数。包括MKL-DNN会用到`MkldnnStream`和`CpuEngine`，和未来可能还会用到`FPGAEngine`等。
5. 在**Argument**里添加两个`MkldnnMatrixPtr`，取名为`mkldnnValue`和`mkldnnGrad`，用于存放`MkldnnLayer`会用到的memory buffer。 并且添加函数cvt(会修改为一个更加合适的函数名)，用于处理"CPU device"和"MKL-DNN device"之间memory的相互转化。
6. 在父类`Layer`中的`getOutput`函数中添加一段逻辑，用于判断`deviceId`，并针对device在MKL-DNN和CPU之间不统一的情况，做一个前期转换。 也就是调用`Argument`的cvt函数把output统一到需要的device上。
7. 在原来的`FLAGS`中添加一个`use_mkldnn`的flag，用于选择是否使用MKL-DNN的相关功能。
8. 关于MKLDNN参数的保存。由于MKLDNN参数的格式与PaddlePaddle原有的格式存在不一样的情况，所以需要在保存参数时同时保存该格式信息。目前准备扩展[Header](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/parameter/Parameter.h#L247)里面的`int32_t version`。这个值不管是在v1还是在v2里面，一直保存的是0，所以可以充分利用这个信息，定义一个枚举处理所有MKLDNN的参数格式，从而`MKLDNNLayer`就可以从输入的参数中获取需要的格式信息。
3. 创建`MKLDNNMatrix`，同时继承`CpuMatrix`和`mkldnn::memory`。用于管理MKL-DNN会用到的相关memory函数、接口以及会用的到格式信息。
4. 创建`MKLDNNBase`，定义一些除了layer和memory相关的类和函数。包括MKL-DNN会用到`MKLDNNStream`和`CPUEngine`，和未来可能还会用到`FPGAEngine`等。
5. 每个`MKLDNNlayer`都会有`inVal_`,`inGrad_`,`outVal_`和`outGrad_`，分别代表input value， input gradient，output value和output gradient。他们会存放MKL-DNN用到的internal memory。同时还会定义以*ext*开头的`MKLDNNMatrix`(表示external的memory)，主要是在格式与PaddlePaddle默认的`nchw`格式不匹配时，用于转换内存的工作。必要的转换函数也会在`MKLDNNLayer`中提前定义好，每个子类只需要调用定义好的reset buffer函数即可。
6. 每个`MKLDNNlayer`的resetbuffer相关的函数（包括reset input、output的Value和grad），他们会根据输入参数reset internal和external的memory，当然这两者也可以相等，即表示不需要转换。只需要把握一个原则，每个`MKLDNNlayer`的子类，只需要使用internal的memory就可以了，所有external的转换工作在父类的reset函数中都提前准备好了。
7. 一般来说，external的memory会尽量与PaddlePaddle中的`value`和`grad`共享内存。同时每个`MKLDNNLayer`中的external output value和gradient(也就是`extOutVal_`和`extOutGrad_`)必须分别与`output_.value`和`output_.grad`共享内存，因为PaddlePaddle的activation会直接使用`output_.value`和`output_.grad`。如果不需要external的buffer用于转换，那么internal的buffer也会与他们共享内存。
8. 如果MKL-DNN layer的后面接有cpu device，那么就会使`output_.value`与`extOutVal_`共享内存，同时数据格式就是`nchw`，这样下一个cpu device就能拿到正确的数据。在有cpu device的时候，external的memory的格式始终是`nchw`或者`nc`。
9. 由于MKL-DNN的输出操作都是覆盖data的，不是在原来的数据上累加，所以当网络出现分支时，在`backward`时会需要merge不同layer的梯度。`MKLDNNlayer`中会实现merge的方法，此时每个小分支的input gradient会先临时保存在一个`MKLDNNMatrix`中，由分支处的layer负责求和，并把结果放到这个layer的`output_.grad`中。所以整体上，每个子类并不会需要关心分支的事情，也是在父类都实现好了。
10. 在原来的`FLAGS`中添加一个`use_mkldnn`的flag，用于选择是否使用MKL-DNN的相关功能。
## References
......
doc/design/model_format.md 0 → 100644
浏览文件 @ 06c7686a
# Design Doc: Model Format
## Motivation
A model is an output of the training process. One complete model consists of two parts, the **topology** and the **parameters**. In order to support industrial deployment, the model format must be self-complete and must not expose any training source code.
As a result, In PaddlePaddle, the **topology** is represented as a [ProgramDesc](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/doc/design/program.md), which describes the model structure. The **parameters** contain all the trainable weights in the model. We must support large size parameters and efficient serialization/deserialization of parameters.
## Implementation
The topology is saved as a plain text in a detailed self-contain protobuf file.
The parameters are saved as a binary file. As we all know, the protobuf message has a limit of [64M size](https://developers.google.com/protocol-buffers/docs/reference/cpp/google.protobuf.io.coded_stream#CodedInputStream.SetTotalBytesLimit.details). We have done a [benchmark experiment](https://github.com/PaddlePaddle/Paddle/pull/4610), which shows that protobuf is not fit for the task.
As a result, we design a particular format for tensor serialization. By default, an arbitrary tensor in Paddle is a [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md), and has a description information proto of [LoDTensorDesc](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L99). We save the DescProto as the byte string header. It contains all the necessary information, such as the `dims`, and the `LoD` information in [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/1c0a4c901c9fc881d120249c703b15d1c50dae7d/paddle/framework/lod_tensor.md). A tensor stores values in a continuous memory buffer. For speed we dump the raw memory to disk and save it as the byte string content. So, the binary format of one tensor is,
The table below shows a tensor's byte view in detail. Note that all the signed values are written in the little-endian format.
|field name | type | description |
| --- | --- | --- |
| version | uint32_t | Version of saved file. Always 0 now. |
| tensor desc length | uint32_t | TensorDesc(Protobuf message) length in bytes. |
| tensor desc | void* | TensorDesc protobuf binary message |
| tensor data | void* | Tensor's data in binary format. The length of `tensor_data` is decided by `TensorDesc.dims()` and `TensorDesc.data_type()` |
| lod_level | uint64_t | Level of LoD |
| length of lod[0] | uint64_t | [Optional] length of lod[0] in bytes. |
| data of lod[0] | uint64_t* | [Optional] lod[0].data() |
| ... | ... | ... |
## Summary
- We introduce a model format.
- The model represented by its forward-pass computation procedure is saved in a **ProgramDesc** protobuf message.
- A bunch of specified format binary tensors describe the **parameters**.
doc/design/ops/sequence_decoder.md 0 → 100644
浏览文件 @ 06c7686a
# Design: Sequence Decoder Generating LoDTensors
In tasks such as machine translation and image to text,
a [sequence decoder](https://github.com/PaddlePaddle/book/blob/develop/08.machine_translation/README.md) is necessary to generate sequences.
This documentation describes how to implement the sequence decoder as an operator.
## Beam Search based Decoder
The [beam search algorithm](https://en.wikipedia.org/wiki/Beam_search) is necessary when generating sequences,
it is a heuristic search algorithm that explores the paths by expanding the most promising node in a limited set.
In the old version of PaddlePaddle, a C++ class `RecurrentGradientMachine` implements the general sequence decoder based on beam search,
due to the complexity, the implementation relays on a lot of special data structures,
quite trivial and hard to be customized by users.
There are a lot of heuristic tricks in the sequence generation tasks,
so the flexibility of sequence decoder is very important to users.
During PaddlePaddle's refactoring work,
some new concept is proposed such as [LoDTensor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/lod_tensor.md) and [TensorArray](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/tensor_array.md) that can better support sequence usage,
and they can help to make the implementation of beam search based sequence decoder **more transparent and modular** .
For example, the RNN sates, candidates IDs and probabilities of beam search can be represented as `LoDTensors`;
the selected candidate's IDs in each time step can be stored in a `TensorArray`, and `Packed` to the sentences translated.
## Changing LoD's absolute offset to relative offsets
The current `LoDTensor` is designed to store levels of variable-length sequences,
it stores several arrays of integers each represents a level.
The integers in each level represents the begin and end (not inclusive) offset of a sequence **in the underlying tensor**,
let's call this format the **absolute-offset LoD** for clear.
The relative-offset LoD can fast retrieve any sequence but fails to represent empty sequences, for example, a two-level LoD is as follows
```python
[[0, 3, 9]
[0, 2, 3, 3, 3, 9]]
```
The first level tells that there are two sequences:
- the first's offset is `[0, 3)`
- the second's offset is `[3, 9)`
while on the second level, there are several empty sequences that both begin and end at `3`.
It is impossible to tell how many empty second-level sequences exist in the first-level sequences.
There are many scenarios that relay on empty sequence representation,
such as machine translation or image to text, one instance has no translations or the empty candidate set for a prefix.
So let's introduce another format of LoD,
it stores **the offsets of the lower level sequences** and is called **relative-offset** LoD.
For example, to represent the same sequences of the above data
```python
[[0, 3, 6]
[0, 2, 3, 3, 3, 9]]
```
the first level represents that there are two sequences,
their offsets in the second-level LoD is `[0, 3)` and `[3, 5)`.
The second level is the same with the relative offset example because the lower level is a tensor.
It is easy to find out the second sequence in the first-level LoD has two empty sequences.
The following demos are based on relative-offset LoD.
## Usage in a simple machine translation model
Let's start from a simple machine translation model that is simplified from [machine translation chapter](https://github.com/PaddlePaddle/book/tree/develop/08.machine_translation) to draw a simple blueprint of what a sequence decoder can do and how to use it.
The model has an encoder that learns the semantic vector from a sequence,
and a decoder which uses the sequence decoder to generate new sentences.
**Encoder**
```python
import paddle as pd
dict_size = 8000
source_dict_size = dict_size
target_dict_size = dict_size
word_vector_dim = 128
encoder_dim = 128
decoder_dim = 128
beam_size = 5
max_length = 120
# encoder
src_word_id = pd.data(
name='source_language_word',
type=pd.data.integer_value_sequence(source_dict_dim))
src_embedding = pd.embedding(size=source_dict_size, size=word_vector_dim)
src_word_vec = pd.lookup(src_embedding, src_word_id)
encoder_out_seq = pd.gru(input=src_word_vec, size=encoder_dim)
encoder_ctx = pd.last_seq(encoder_out_seq)
# encoder_ctx_proj is the learned semantic vector
encoder_ctx_proj = pd.fc(
encoder_ctx, size=decoder_dim, act=pd.activation.Tanh(), bias=None)
```
**Decoder**
```python
def generate():
decoder = pd.while_loop()
with decoder.step():
decoder_mem = decoder.memory(init=encoder_ctx) # mark the memory
generated_ids = decoder.memory() # TODO init to batch_size <s>s
generated_scores = decoder.memory() # TODO init to batch_size 1s or 0s
target_word = pd.lookup(trg_embedding, gendrated_ids)
# expand encoder_ctx's batch to fit target_word's lod
# for example
# decoder_mem.lod is
# [[0 1 3],
# [0 1 3 6]]
# its tensor content is [a1 a2 a3 a4 a5]
# which means there are 2 sentences to translate
# - the first sentence has 1 translation prefixes, the offsets are [0, 1)
# - the second sentence has 2 translation prefixes, the offsets are [1, 3) and [3, 6)
# the target_word.lod is
# [[0, 1, 6]
# [0, 2, 4, 7, 9 12]]
# which means 2 sentences to translate, each has 1 and 5 prefixes
# the first prefix has 2 candidates
# the following has 2, 3, 2, 3 candidates
# the encoder_ctx_expanded's content will be
# [a1 a1 a2 a2 a3 a3 a3 a4 a4 a5 a5 a5]
encoder_ctx_expanded = pd.lod_expand(encoder_ctx, target_word)
decoder_input = pd.fc(
act=pd.activation.Linear(),
input=[target_word, encoder_ctx],
size=3 * decoder_dim)
gru_out, cur_mem = pd.gru_step(
decoder_input, mem=decoder_mem, size=decoder_dim)
scores = pd.fc(
gru_out,
size=trg_dic_size,
bias=None,
act=pd.activation.Softmax())
# K is an config
topk_scores, topk_ids = pd.top_k(scores, K)
topk_generated_scores = pd.add_scalar(topk_scores, generated_scores)
selected_ids, selected_generation_scores = decoder.beam_search(
topk_ids, topk_generated_scores)
# update the states
decoder_mem.update(cur_mem) # tells how to update state
generated_ids.update(selected_ids)
generated_scores.update(selected_generation_scores)
decoder.output(selected_ids)
decoder.output(selected_generation_scores)
translation_ids, translation_scores = decoder()
```
The `decoder.beam_search` is a operator that given the candidates and the scores of translations including the candidates,
return the result of the beam search algorithm.
In this way, users can customize anything on the inputs or outputs of beam search, for example, two ways to prune some translation prefixes
1. meke the correspondind elements in `topk_generated_scores` zero or some small values, beam_search will discard this candidate.
2. remove some specific candidate in `selected_ids`
3. get the final `translation_ids`, remove the translation sequence in it.
The implementation of sequence decoder can reuse the C++ class [RNNAlgorithm](https://github.com/Superjom/Paddle/blob/68cac3c0f8451fe62a4cdf156747d6dc0ee000b3/paddle/operators/dynamic_recurrent_op.h#L30),
so the python syntax is quite similar to a [RNN](https://github.com/Superjom/Paddle/blob/68cac3c0f8451fe62a4cdf156747d6dc0ee000b3/doc/design/block.md#blocks-with-for-and-rnnop).
Both of them are two-level `LoDTensors`
- the first level represents `batch_size` of (source) sentences;
- the second level represents the candidate ID sets for translation prefix.
for example, 3 source sentences to translate, and has 2, 3, 1 candidates.
Unlike an RNN, in sequence decoder, the previous state and the current state have different LoD and shape,
a `lod_expand` operator is used to expand the LoD of the previous state to fit the current state.
For example, the previous state
* LoD is `[0, 1, 3][0, 2, 5, 6]`
* content of tensor is `a1 a2 b1 b2 b3 c1`
the current state stored in `encoder_ctx_expanded`
* LoD is `[0, 2, 7][0 3 5 8 9 11 11]`
* the content is
- a1 a1 a1 (a1 has 3 candidates, so the state should be copied 3 times for each candidates)
- a2 a2
- b1 b1 b1
- b2
- b3 b3
- None (c1 has 0 candidates, so c1 is dropped)
Benefit from the relative offset LoD, empty candidate set can be represented naturally.
the status in each time step can be stored in `TensorArray`, and `Pack`ed to a final LoDTensor, the corresponding syntax is
```python
decoder.output(selected_ids)
decoder.output(selected_generation_scores)
```
the `selected_ids` is the candidate ids for the prefixes,
it will be `Packed` by `TensorArray` to a two-level `LoDTensor`,
the first level represents the source sequences,
the second level represents generated sequences.
Pack the `selected_scores` will get a `LoDTensor` that stores scores of each candidate of translations.
Pack the `selected_generation_scores` will get a `LoDTensor`, and each tail is the probability of the translation.
## LoD and shape changes during decoding
<p align="center">
<img src="./images/LOD-and-shape-changes-during-decoding.jpg"/>
</p>
According the image above, the only phrase to change LoD is beam search.
## Beam search design
The beam search algorthm will be implemented as one method of the sequence decoder, it has 3 inputs
1. `topk_ids`, top K candidate ids for each prefix.
2. `topk_scores`, the corresponding scores for `topk_ids`
3. `generated_scores`, the score of the prefixes.
All of the are LoDTensors, so that the sequence affilication is clear.
Beam search will keep a beam for each prefix and select a smaller candidate set for each prefix.
It will return three variables
1. `selected_ids`, the final candidate beam search function selected for the next step.
2. `selected_scores`, the scores for the candidates.
3. `generated_scores`, the updated scores for each prefixes (with the new candidates appended).
## Introducing the LoD-based `Pack` and `Unpack` methods in `TensorArray`
The `selected_ids`, `selected_scores` and `generated_scores` are LoDTensors,
and they exist in each time step,
so it is natural to store them in arrays.
Currently, PaddlePaddle has a module called `TensorArray` which can store an array of tensors,
the results of beam search are better to store in a `TensorArray`.
The `Pack` and `UnPack` in `TensorArray` are used to package tensors in the array to a `LoDTensor` or split the `LoDTensor` to an array of tensors.
It needs some extensions to support pack or unpack an array of `LoDTensors`.
doc/design/optimizer.md
浏览文件 @ 06c7686a
......@@ -65,20 +65,6 @@ class Optimizer(object):
def __init__(self):
pass
def create_backward_pass(self, loss, parameter_list=None):
"""
create and add gradient Operators in BlockDesc to Compute gradients of `loss`
for parameters in parameter_list
Args:
loss: an variable generated by cost function.
parameter_list: parameters that need to compute gradient and update to optimize the lost.
Returns:
list of (parameters, gradients) pair.
"""
return None
def create_optimization_pass(self, parameters_and_grads):
"""Add optimization operators to update gradients to variables.
......@@ -93,7 +79,7 @@ class Optimizer(object):
def minimize(self, loss, parameter_list):
"""Add operations to minimize `loss` by updating `parameter_list`.
This method combines interface `create_backward_pass()` and
This method combines interface `append_backward_ops()` and
`create_optimization_pass()` into one.
"""
params_grads = self.create_backward_pass(loss, parameter_list)
......
doc/design/parameter_average.md 0 → 100644
浏览文件 @ 06c7686a
# Averaging Parameter in PaddlePaddle
## Why Averaging
In a large scale machine learning setup where the size of the training data is huge, it could take us a large number of iterations over the training data before we can achieve the optimal values of parameters of our model. Looking at the problem setup, it is desirable if we can obtain the optimal values of parameters by going through the data in as few passes as we can.
Polyak and Juditsky (1992) showed that the test performance of simple average of parameters obtained by Stochastic Gradient Descent (SGD) is as good as that of parameter values that are obtained by training the model over and over again, over the training dataset.
Hence, to accelerate the speed of Stochastic Gradient Descent, Averaged Stochastic Gradient Descent (ASGD) was proposed in Polyak and Juditsky (1992). For ASGD, the running average of parameters obtained by SGD, is used as the estimator for <img src="./images/theta_star.gif"/><br/> . The averaging is done as follows:
<img src="./images/asgd.gif" align="center"/><br/>
We propose averaging for any optimizer similar to how ASGD performs it, as mentioned above.
### How to perform Parameter Averaging in PaddlePaddle
Parameter Averaging in PaddlePaddle works in the following way during training :
1. It will take in an instance of a normal optimizer as an input, e.g. RMSPropOptimizer
2. The optimizer itself is responsible for updating the parameters.
3. The ParameterAverageOptimizer maintains a separate copy of the parameters for itself:
1. In concept, the values of this copy are the average of the values of the parameters in the most recent N batches.
2. However, saving all the N instances of the parameters in memory is not feasible.
3. Therefore, an approximation algorithm is used.
Hence, overall we have have two copies of the parameters: one for the optimizer itself, and one for the ParameterAverageOptimizer. The former should be used in back propagation, while the latter should be used during testing and should be saved.
During the testing/ saving the model phase, we perform the following steps:
1. Perform the delayed operations.
2. Save current values of the parameters to a temporary variable.
3. Replace the values of the parameters with the averaged values.
4. Perform testing and/or save the parameters.
5. Restore the values of the parameters once done.
### How to implement Averaging of Parameter in PaddlePaddle
We can add the ParameterAverageOptimizer op to the graph through Python API. Using this approach, we manually add this op to the graph and direct the output of the optimizer op to this op during training.
**Advantages**:
- Allows for greater flexibility to the users of PaddlePaddle. Using this approach, the users can plug different optimizers into ParameterAverageOptimizer by passing in the optimizer to the op.
- Makes it easy for the users to customize and extend the framework.
**Disadvantages**:
- Implementation requires re-writing the averaging methodology in Python.
### Low-Level implementation
In the new design, we propose to create a new operation for averaging parameter updates (ParameterAverageOptimizer). For now, we can add an op that takes in the following as input:
- the optimizer
- the window_size to keep the updates
The ParameterAverageOptimizer op can be like any other operator with its own CPU/GPU implementation either using Eigen or separate CPU and GPU kernels. As the initial implementation, we can implement the kernel using Eigen following the abstraction pattern implemented for [Operators](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/rmsprop_op.h). We also want to support the case when the Trainer/Optimizer runs on the GPU while ParameterAverageOptimizer runs on a CPU.
The idea of building an op for averaging is in sync with the refactored PaddlePaddle philosophy of using operators to represent any computation unit. The way the op will be added to the computation graph will be decided by the [layer functions](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md#layer-function) in Python API.
### Python API implementation for ParameterAverageOptimizer
Based on Polyak and Juditsky (1992), we can generalize the averaging of updates to any optimizer. The input to the op would be the following:
- Any optimizer (RMSProp , AdaGrad etc.)
- A window size. The op keeps accumulating updated parameter values over a window of N batches and takes an average. Move the averaged value to a buffer when window is full to avoid loss of precision.
Using the ParameterAverageOptimizer op, any user can add the operation to their computation graphs. However, this will require a lot of lines of code and we should design Python APIs that support averaging. As per the PaddlePaddle [Python API design](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md), the layer functions are responsible for creating operators, operator parameters and variables. Since ParameterAverageOptimizer will be an operator, it makes sense to create it in the layer functions.
We will have a wrapper written in Python that will support the functionality and implement the actual core computation in C++ core as we have done for other [Optimizers](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/rmsprop_op.cc)
#### Creation of the ParameterAverageOptimizer operator
There are two ways for creating the ParameterAverageOptimizer op:
1. We create the op immediately while building the computation graph.
2. We add the op in a lazy manner, just before the backward pass, similar to the way the optimization ops are added.
The proposal is to add the op immediately while building the computation graph.
#### High-level API
In PaddlePaddle Python API, users will primarily rely on [layer functions](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md#layer-function) to create neural network layers. Hence, we also need to provide parameter average functionality in layer functions.
doc/design/prune.md 0 → 100644
浏览文件 @ 06c7686a
# Prune
## Motivation
We want to support running inference, training and checkpointing in one `ProgramDesc`. We implement
`void Prune(const ProgramDesc* input, ProgramDesc* output)` function, which takes a `ProgramDesc`
and generate a pruned `ProgramDesc`.
## Challenge
Pruning need to support both variables and operators being evaluation targets. Consider the following
different situations.
```python
# Case 1: run foward pass.
cost_np = session.run(target=cost)
# Case 2: run backward passing.
opts_np, _ = session.run(target=[cost, opt])
# Case 3: run checkpointing
_ = session.run(target=checkpoint)
```
## Solution
To support evaluation of operators, we add `is_target` field in the `OpDesc`.
```c++
message OpDesc {
required string type = 3;
repeated Var inputs = 1;
repeated Var outputs = 2;
repeated Attr attrs = 4;
optional bool is_target = 5 [ default = false ];
};
```
To support evaluation of variables, we add [fetch_op](https://github.com/PaddlePaddle/Paddle/pull/4599).
For each variable in the `target`, we insert a `fetch_op` into the `ProgramDesc` with `variable` being
`fetch_op`'s input. Then we also set `fetch_op` is a target.
### Algorithm
If an operator needs to be run, it must fall into one of the following cases:
1. It is the target.
2. It is depended by some other ops, meaning its output is some other op's input.
The first case can be checked by `op_desc.is_traget()` . The second case can be implement as
```c++
bool HasDependentVar(const OpDesc& op_desc, const std::set<string>& dependent_vars) {
for (auto& var : op_desc.outputs()) {
for (auto& argu : var.arguments()) {
if (dependent_vars.count(argu) != 0) {
return true;
}
}
}
return false;
}
```
Then the whole algorithm can be implemented as the following [code](https://github.com/tonyyang-svail/Paddle/blob/prune_impl/paddle/framework/prune.cc).
doc/design/refactorization.md
浏览文件 @ 06c7686a
......@@ -177,9 +177,6 @@ REGISTER_OP(op_type, op_class, op_maker_class, grad_op_type, grad_op_class)
REGISTER_OP_WITHOUT_GRADIENT(op_type, op_class, op_maker_class)
```
### USE Macros
Make sure the registration process is executed and linked.
---
# Registration Process
1. Write an Op class and its gradient Op class, if required.
......@@ -188,8 +185,6 @@ Make sure the registration process is executed and linked.
1. Call maker class to complete `proto` and `checker`
2. Using the completed `proto` and `checker`, it will add a new key-value pair to the `OpInfoMap`
4. Invoke the `USE` macro in which the Op is used to make sure that it is linked.
---
# Backward Module (1/2)
### Create Backward Operator
......
doc/design/register_grad_op.md
浏览文件 @ 06c7686a
......@@ -3,17 +3,17 @@
## The Problem Posed
Currently, for each C++ operator class definition, there registers a *gradient operator creator* function, which takes a C++ operator instance and returns the corresponding gradient operator instance.
Currently, for each C++ operator class definition, a *gradient operator creator* function is registered, which takes as input a C++ operator instance and returns the corresponding gradient operator instance.
However, we noticed two problems with the current deisgn:
However, we noticed two problems with the current design:
1. As we decided to separate the *compilation* and *execution* phases, we need to change the creator to take an `OpDesc` protobuf message in a `ProgramDesc` and inserts corresponding `OpDesc` messages into the `ProgramDesc` message.
1. As we decided to separate the *compilation* and the *execution* phases, we need to change the creator to take an `OpDesc` protobuf message in a `ProgramDesc` and inserts corresponding `OpDesc` messages into the `ProgramDesc` message.
1. Some operator's gradient computation requires more than one gradient operators. For example, the gradient of *minus* consists of two operators -- an identity operaotr and a scale operator. So we need to make the registration mechanism to support the mapping from an operator to a set of operators for gradient computation.
1. For some operators, the gradient computation can be written in terms of existing operators. For example, the gradient of *minus* operator consists of two operators -- an *identity* operator followed by a *scale* operator. Hence the registration mechanism needs to support mapping from an operator to a set of operators for the gradient computation.
## The Current Implementation
The C++ class `OpInfos` store in a association map which key is the operator type. The `grad_op_type` indicate associated gradient operator type. Operator can create gradient operator by `OpInfo::creator_` of gradient. The pseudo code is
Instances of the C++ class `OpInfo` are stored an associative map whose key is the operator type. The `grad_op_type` indicates the associated gradient operator type. An operator can create the gradient operator by invoking `OpInfo::creator_` of the gradient operator. The pseudo code is as follows
```cpp
struct OpInfo {
......@@ -31,16 +31,16 @@ OperatorBase* CreateGradientOperator(const OperatorBase& op) {
## Proposed Solution
The mapping relationship between an operator and its gradient operators is a function. The interface of that function is:
The mapping relationship between an operator and its gradient operators is a function. The interface of this function is:
```cpp
// (OpDesc) --> vector<OpDesc>
std::function<std::vector<OpDescBind>(const OpDescBind&)>;
```
The function takes an `OpDescBind` of the forward operator and returns one or many gradient operator descriptions. `OpDescBind` is a C++ wrapper for protobuf message `OpDesc` to manipulate `OpDesc` fast.
The function takes an `OpDescBind` of the forward operator and returns one or many gradient operator descriptions. `OpDescBind` is a C++ wrapper for the protobuf message `OpDesc` for rapid manipulation of `OpDesc`.
The `GradOpDescMaker` will be registered in `OpInfo`, to replace `grad_op_type_` field. The `OpInfo` should be
The `GradOpDescMaker` will be registered in `OpInfo` and will replace the `grad_op_type_` field. The `OpInfo` should look like
```cpp
struct OpInfo {
......@@ -49,7 +49,7 @@ struct OpInfo {
};
```
The `grad_op_maker_ ` is `nullptr` if the operator does not have associated gradient operators.
The `grad_op_maker_ ` is a `nullptr` if the operator does not have any associated gradient operators.
We propose a base class called `GradOpDescMakerBase` to let operator developers generate `Gradient Operators` easily. The public interface of that class is
......@@ -74,7 +74,7 @@ func = [] (const OpDescBind& fwd_op) {
We can write many helper functions since the `GradOpDescMakerBase` is a class now. The basic helper functions get the variables of `Input`, `Output`, `InputGradient` and `OutputGradient` in the forwarding operator.
We should chagne register macros at the same time. In the current solution, there is no difference between forwarding operators and backward operators. So `REGISTER_OP` just register one operator. If the `REGISTER_OPERATOR ` contains `OpProtoAndCheckerMaker` and `GradOpDescMaker`, we just list them in the same macro. It can be done by a macro contains `__VA_ARGS__`.
We should change register macros at the same time. In the current solution, there is no difference between forwarding operators and backward operators. So `REGISTER_OP` just register one operator. If the `REGISTER_OPERATOR ` contains `OpProtoAndCheckerMaker` and `GradOpDescMaker`, we just list them in the same macro. It can be done by a macro contains `__VA_ARGS__`.
The user interface should be
......
doc/design/regularization.md 0 → 100644
浏览文件 @ 06c7686a
# Regularization in PaddlePaddle
## Introduction to Regularization
A central problem in machine learning is how to design an algorithm that will perform well not just on the training data, but also on new data. A frequently faced problem is the problem of **overfitting**, where the model does not make reliable predictions on new unseen data. **Regularization** is the process of introducing additional information in order to prevent overfitting. This is usually done by adding extra penalties to the loss function that restricts the parameter spaces that an optimization algorithm can explore.
### Parameter Norm Penalties
Most common regularization approaches in deep learning are based on limiting the capacity of the models by adding a parameter norm penalty to the objective function `J`. This is given as follows:
<img src="./images/loss_equation.png" align="center"/><br/>
The parameter `alpha` is a hyperparameter that weights the relative contribution of the norm penalty term, `omega`, relative to the standard objective function `J`.
The most commonly used norm penalties are the L2 norm penalty and the L1 norm penalty. These are given as follows:
##### L2 Regularization:
<img src="./images/l2_regularization.png" align="center"/><br/>
##### L1 Regularization
<img src="./images/l1_regularization.png" align="center"/><br/>
A much more detailed mathematical background of regularization can be found [here](http://www.deeplearningbook.org/contents/regularization.html).
## Regularization Survey
A detailed survey of regularization in various deep learning frameworks can be found [here](https://github.com/PaddlePaddle/Paddle/wiki/Regularization-Survey).
## Proposal for Regularization in PaddlePaddle
### Low-Level implementation
In the new design, we propose to create new operations for regularization. For now, we can add 2 ops that correspond to the most frequently used regularizations:
- L2_regularization_op
- L1_regularization_op
These ops can be like any other ops with their own CPU/GPU implementations either using Eigen or separate CPU and GPU kernels. As the initial implementation, we can implement their kernels using Eigen following the abstraction pattern implemented for [Activation Ops](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/accuracy_op.h). This abstraction pattern can make it very easy to implement new regularization schemes other than L1 and L2 norm penalties.
The idea of building ops for regularization is in sync with the refactored Paddle philosophy of using operators to represent any computation unit. The way these ops will be added to the computation graph, will be decided by the [layer functions](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md#layer-function) in Python API.
### Computation Graph
Below is an example of a really simple feed forward neural network.
<img src="./images/feed_forward.png" align="center"/><br/>
The Python API will modify this computation graph to add regularization operators. The modified computation graph will look as follows:
<img src="./images/feed_forward_regularized.png" align="center"/><br/>
   
### Python API implementation for Regularization
Using the low level ops, `L2_regularization_op` and `L1_regularization_op`, any user can add regularization to their computation graphs. However, this will require a lot of lines of code and we should design Python APIs that support regularization. An example of such an API can be seen in [Keras](https://keras.io/regularizers/). As per the PaddlePaddle [Python API design](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md), the layer functions are responsible for creating operators, operator parameters and variables. Since regularization is a property of parameters, it makes sense to create these in the layer functions.
#### Creation of Regularization ops
There are two possibilities for creating the regularization ops:
1. We create these ops immediately while building the computation graph.
2. We add these ops in a lazy manner, just before the backward, similar to the way the optimization ops are added.
The proposal is to add these ops in a lazy manner just before the backward pass.
#### Storage of Regularization attributes
Since we want to create the regularization ops in a lazy manner, the regularization attributes (type of regularization and weight of regularization penalty) can be stored as attributes of the [`Parameter`](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/framework/framework.py#L421) class. This is because regularization is a property of the parameters and storing regularization properties with Parameters also allows for shared parameters.
#### High-level API
In PaddlePaddle Python API, users will primarily rely on [layer functions](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/python_api.md#layer-function) to create neural network layers. Hence, we also need to provide regularization functionality in layer functions. The design of these APIs can be postponed for later right now. A good reference for these APIs can be found in [Keras](https://keras.io/regularizers/) and also by looking at Tensorflow in [`tf.contrib.layers`](https://www.tensorflow.org/api_guides/python/contrib.layers).
doc/design/selected_rows.md
浏览文件 @ 06c7686a
# Design Doc: Selected Rows
`SelectedRows` is a kind of sparse tensor data type, which is designed to support `embedding` operators. The gradient of embedding table is a sparse tensor. Only a few rows are non-zero values in that tensor. It is straightforward to represent the sparse tensor by the following sparse tensor data structure:
`SelectedRows` is a type of sparse tensor data type, which is designed to support `embedding` operators. The gradient of embedding table is a sparse tensor. Only a few rows are non-zero values in this tensor. It is straight-forward to represent a sparse tensor by the following sparse tensor data structure:
```cpp
class SelectedRows {
......@@ -11,7 +11,7 @@ class SelectedRows {
};
```
The field `height_` shows the first dimension of `SelectedRows`. The `rows` are the indices of which rows of `SelectedRows` are non-zeros. The `value_` field is an N-dim tensor and shape is `[rows.size() /* NUM_ROWS */, ...]`, which supplies values for each row. The dimension of `SelectedRows` satisfies `[height_] + value_.shape[1:]`.
The field `height_` is the first dimension of `SelectedRows`. The `rows` are the indices of the non-zero rows of `SelectedRows`. The `value_` field is an N-dim tensor of shape `[rows.size() /* NUM_ROWS */, ...]`, which supplies values for each row. The dimension of `SelectedRows` satisfies `[height_] + value_.shape[1:]`.
Suppose that a SelectedRows-typed variable `x` has many rows, but only two of them have values -- row 73 is `[1, 2]` and row 84 is `[3, 4]`, the `SelectedRows` representation would be:
......@@ -25,7 +25,7 @@ x = SelectedRow {
## SelectedRows in Protobuf
`SelectedRows` is a kind of `Variable`. `VarDesc` in protobuf should describe the `SelectedRows` information. Only the tensor dimension of a `SelectedRows` will be described in compile-time since the `rows_` and `value_` are related to training data.
`SelectedRows` is a type of `Variable`. `VarDesc` in protobuf should describe the `SelectedRows` information. Only the tensor dimension of a `SelectedRows` will be described in compile-time because the `rows_` and `value_` are dependent on the training data.
So we use `TensorDesc` to unify `data_type` and `dims`. A LodTensorDesc contains a `TensorDesc` and `lod_level`. The description of `SelectedRows` is a Tensor description.
```proto
......@@ -54,7 +54,7 @@ message VarDesc {
## InferShape for Selected Rows
Just like `LoD` information, `InferShape` method will inference output tensor type as well. The operator should decide whether its output is a `SelectedRows` or `Dense` tensor.
Just like `LoD` information, `InferShape` method will infer the output tensor type as well. The operator should decide whether its output is a `SelectedRows` or `Dense` tensor.
For example, the gradient operator of `TableLookup` will always generate `SelectedRows`. Its `InferShape` method should be like following
......@@ -68,7 +68,7 @@ void TableLookupGrad::InferShape(context) {
## Sparse Operators
There are several operators should be written to support `SelectedRows`. They are:
There are several operators that need to be written to support `SelectedRows`. These are:
1. Operators which generates `SelectedRows` gradient. e.g. Gradient of `TableLookupOp`.
1. Operators which generate `SelectedRows` gradient. e.g. Gradient of `TableLookupOp`.
2. Optimize operators which support `SelectedRows` gradient. e.g. `SGD` or `AdaGrad` for `SelectedRows`. However, there should be only one `SGD` operator. `OpWithKernel::Run` should select a suitable kernel for both `dense` tensor or `SelectedRows`.
doc/faq/local/index_cn.rst
浏览文件 @ 06c7686a
......@@ -99,7 +99,7 @@ PaddlePaddle支持Sparse的训练，sparse训练需要训练特征是 :code:`spa
利用更多的计算资源
++++++++++++++++++
利用更多的计算资源可以分为一下几个方式来进行\:
利用更多的计算资源可以分为以下几个方式来进行\:
* 单机CPU训练
......@@ -174,7 +174,7 @@ decoder_inputs = paddle.layer.fc(
1. 两者都是对梯度的截断，但截断时机不同，前者在 :code:`optimzier` 更新网络参数时应用；后者在激活函数反向计算时被调用；
2. 截断对象不同：前者截断可学习参数的梯度，后者截断回传给前层的梯度;
除此之外，还可以通过减小学习律或者对数据进行归一化处理来解决这类问题。
除此之外，还可以通过减小学习率或者对数据进行归一化处理来解决这类问题。
5. 如何调用 infer 接口输出多个layer的预测结果
-----------------------------------------------
......
doc/faq/parameter/index_cn.rst
浏览文件 @ 06c7686a
......@@ -75,7 +75,7 @@ PaddlePaddle目前支持8种learning_rate_schedule，这8种learning_rate_schedu
optimizer = paddle.optimizer.Adam(
learning_rate=1e-3,
learning_rate_schedule="manual",
learning_rate_schedule="pass_manual",
learning_rate_args="1:1.0,2:0.9,3:0.8",)
在该示例中，当已训练pass数小于等于1时，学习率为 :code:`1e-3 * 1.0`；当已训练pass数大于1小于等于2时，学习率为 :code:`1e-3 * 0.9`；当已训练pass数大于2时，学习率为 :code:`1e-3 * 0.8`。
......
doc/getstarted/build_and_install/docker_install_cn.rst
浏览文件 @ 06c7686a
......@@ -145,7 +145,7 @@ PaddlePaddle发布新版本的时候都会发布对应版本的生产镜像以
Jupyter Notebook是一个开源的web程序，大家可以通过它制作和分享带有代码、公式、图表、文字的交互式文档。用户可以通过网页浏览文档。
PaddlePaddle Book是为用户和开发者制作的一个交互式的Jupyter Nodebook。
PaddlePaddle Book是为用户和开发者制作的一个交互式的Jupyter Notebook。
如果您想要更深入了解deep learning，PaddlePaddle Book一定是您最好的选择。
我们提供可以直接运行PaddlePaddle Book的Docker镜像，直接运行：
......
doc/howto/dev/contribute_to_paddle_en.md 已删除 100644 → 0
浏览文件 @ 7461b359
# Contribute Code
We sincerely appreciate your contributions. You can use fork and pull request
workflow to merge your code.
## Code Requirements
- Your code comments must be fully documented by
[Doxygen](http://www.stack.nl/~dimitri/doxygen/) style.
- Make sure the compiler option `WITH_STYLE_CHECK` is on and the compiler
passes the code style check.
- All code must have unit test.
- Pass all unit tests.
The following tutorial guides you into submitting your contibution.
## [Creating a Fork](https://help.github.com/articles/fork-a-repo/)
Just head over to the GitHub page and click the "Fork" button.
It's just that simple.
## Clone
Clone remote repository.
```bash
➜ git clone https://github.com/USERNAME/Paddle
➜ cd Paddle
```
## Create a local branch
Paddle is currently using [Git-flow branching model](http://nvie.com/posts/a-successful-git-branching-model/).
All feature and bug fix development work should be done on a new branch, generally create new branch from `develop` branch .
```bash
➜ git checkout -b my-cool-stuff
```
Before the checkout, you need to keep the current branch directory clean, otherwise the untracked file will be brought to the new branch, which can be inspected by `git status`.
## Using `pre-commit` hook
Paddle developers use [pre-commit](http://pre-commit.com/) tool to manage git
pre-commit hooks. It can help us format source codes (cpp, python), check some
basic thing before commit (only one EOL for each file, do not add a huge file
in git). `pre-commit` tests is a part of unit tests in Travis-CI now, every
PR doesn't fit hook can not be merged into Paddle.
To use [pre-commit](http://pre-commit.com/), you should install it by
`pip install pre-commit`, and currently, Paddle uses `clang-format` to format
c/cpp sources. Please make sure clang-format 3.8+ installed.
Install and run it as follow:
```bash
➜ pip install pre-commit
➜ pre-commit install
```
When you commit your code, the pre-commit hook will check the local code if there is
anything not suitable to commit, and so on.
## Start to develop
In this tutorial, I delete a line in README.md and created a new file.
We can use `git status` to inspect the changes of current directory, `git diff` to see difference.
```bash
➜ git status
On branch test
Changes not staged for commit:
(use "git add <file>..." to update what will be committed)
(use "git checkout -- <file>..." to discard changes in working directory)
modified: README.md
Untracked files:
(use "git add <file>..." to include in what will be committed)
test
no changes added to commit (use "git add" and/or "git commit -a")
```
## Build and Test
We package PaddlePaddle's compile environment into a Docker image, called the develop image named `paddle:dev`, it contains all compiling tools that PaddlePaddle needs.
If you want to build the develop image, just run:
```bash
➜ docker build -t paddle:dev .
```
Then we can use the develop image to build PaddlePaddle source. For example:
```bash
➜ docker run -v $(pwd):/paddle -e "WITH_GPU=OFF" -e "WITH_AVX=ON" -e "WITH_TEST=ON" paddle:dev
```
The above command will compile PaddlePaddle and create a Dockerfile for building production image. All the generated files are in the build directory. "WITH_GPU" controls if the generated production image supports GPU. "WITH_AVX" controls if the generated production image supports AVX. "WITH_TEST" controls if the unit test will be generated.
Then we can generate the production image by copying the compiled PaddlePaddle program into the image by
```bash
➜ docker build -t paddle:prod -f build/Dockerfile .
```
Run unit test finally:
```bash
➜ docker run -it -v $(pwd):/paddle paddle:dev bash -c "cd /paddle/build && ctest"
```
For more details, you can read [this doc](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/getstarted/build_and_install/docker_install_en.rst).
## Commit
Next we cancel the changes to the README.md file and then commit our changes by following command lines:
```bash
➜ git checkout -- README.md
➜ git status
On branch test
Untracked files:
(use "git add <file>..." to include in what will be committed)
test
nothing added to commit but untracked files present (use "git add" to track)
➜ git add test
```
We should write a description of each commit by `git commit` to allow others to know
the changes in these files.
```bash
➜ git commit
CRLF end-lines remover...............................(no files to check)Skipped
yapf.................................................(no files to check)Skipped
Check for added large files..............................................Passed
Check for merge conflicts................................................Passed
Check for broken symlinks................................................Passed
Detect Private Key...................................(no files to check)Skipped
Fix End of Files.....................................(no files to check)Skipped
clang-formater.......................................(no files to check)Skipped
[my-cool-stuff c703c041] add test file
1 file changed, 0 insertions(+), 0 deletions(-)
create mode 100644 233
```
## Keeping Fork Up to Date
Before pull your request, you should sync your code from the latest PaddlePaddle.
To do this, you'll need to add a remote at first:
```bash
➜ git remote add upstream https://github.com/PaddlePaddle/Paddle
➜ git remote
origin
upstream
```
Update your fork with the latest upstream changes:
```bash
➜ git fetch upstream
➜ git pull upstream develop
```
Now, your local master branch is up-to-date with everything modified upstream.
## Push to GitHub
```bash
# push to your repository in Github
➜ git push origin my-cool-stuff
```
## Create an issue and a Pull Request
Create an Issue to describe the problem and record its number.
Go to the page for your fork on GitHub, select your development branch,
and click the `New pull request`.
<img width="295" alt="screen shot 2017-04-26 at 9 09 28 pm" src="https://cloud.githubusercontent.com/assets/11692045/25436054/a6d98c66-2ac4-11e7-9cb1-18dd13150230.png">
Then select the target branch:
<img width="750" alt="screen shot 2017-04-26 at 9 11 52 pm" src="https://cloud.githubusercontent.com/assets/11692045/25436139/f83b1e6c-2ac4-11e7-8c0e-add499023c46.png">
We can add `resolve #Issue number` in PR description to close the issue automatically after the PR is merge. More details in <https://help.github.com/articles/closing-issues-via-commit-messages/>.
Then wait for review, if there need to modify, refer to the above steps to update the corresponding origin branch.
## Delete origin branch
After the PR is merge into the main repository, we can delete the remote branch on the PR page.
<img width="775" alt="screen shot 2017-04-26 at 9 18 24 pm" src="https://cloud.githubusercontent.com/assets/11692045/25436457/e4cdd472-2ac5-11e7-9272-badc76c4a23e.png">
Or just run:
```bash
➜ git push origin :my-cool-stuff
```
## Delete local branch
Finally, we delete local branch:
```bash
➜ git checkout develop
# delete my-cool-stuff branch
➜ git branch -D my-cool-stuff
```
doc/howto/dev/contribute_to_paddle_en.md 0 → 120000
浏览文件 @ 06c7686a
../../../CONTRIBUTING.md
\ No newline at end of file
doc/howto/dev/new_op_cn.md
浏览文件 @ 06c7686a
......@@ -214,7 +214,7 @@ MulOp(const std::string &type, const framework::VariableNameMap &inputs,
```cpp
// if use Eigen unsupported module before include head files
#define EIGEN_USE_GPU
// #define EIGEN_USE_GPU
namespace ops = paddle::operators;
REGISTER_OP_GPU_KERNEL(mul, ops::MulKernel<paddle::platform::GPUPlace, float>);
......
doc/howto/index_cn.rst
浏览文件 @ 06c7686a
......@@ -21,7 +21,6 @@
dev/build_cn.rst
dev/write_docs_cn.rst
dev/contribute_to_paddle_cn.md
模型配置
--------
......
doc/howto/usage/cluster/src/word2vec/api_train_v2.py 0 → 100644
浏览文件 @ 06c7686a
import gzip
import math
import paddle.v2 as paddle
embsize = 32
hiddensize = 256
N = 5
def wordemb(inlayer):
wordemb = paddle.layer.embedding(
input=inlayer,
size=embsize,
param_attr=paddle.attr.Param(
name="_proj",
initial_std=0.001,
learning_rate=1,
l2_rate=0,
sparse_update=True))
return wordemb
def main():
# for local training
cluster_train = False
if not cluster_train:
paddle.init(use_gpu=False, trainer_count=1)
else:
paddle.init(
use_gpu=False,
trainer_count=2,
port=7164,
ports_num=1,
ports_num_for_sparse=1,
num_gradient_servers=1)
word_dict = paddle.dataset.imikolov.build_dict()
dict_size = len(word_dict)
firstword = paddle.layer.data(
name="firstw", type=paddle.data_type.integer_value(dict_size))
secondword = paddle.layer.data(
name="secondw", type=paddle.data_type.integer_value(dict_size))
thirdword = paddle.layer.data(
name="thirdw", type=paddle.data_type.integer_value(dict_size))
fourthword = paddle.layer.data(
name="fourthw", type=paddle.data_type.integer_value(dict_size))
nextword = paddle.layer.data(
name="fifthw", type=paddle.data_type.integer_value(dict_size))
Efirst = wordemb(firstword)
Esecond = wordemb(secondword)
Ethird = wordemb(thirdword)
Efourth = wordemb(fourthword)
contextemb = paddle.layer.concat(input=[Efirst, Esecond, Ethird, Efourth])
hidden1 = paddle.layer.fc(input=contextemb,
size=hiddensize,
act=paddle.activation.Sigmoid(),
layer_attr=paddle.attr.Extra(drop_rate=0.5),
bias_attr=paddle.attr.Param(learning_rate=2),
param_attr=paddle.attr.Param(
initial_std=1. / math.sqrt(embsize * 8),
learning_rate=1))
predictword = paddle.layer.fc(input=hidden1,
size=dict_size,
bias_attr=paddle.attr.Param(learning_rate=2),
act=paddle.activation.Softmax())
def event_handler(event):
if isinstance(event, paddle.event.EndIteration):
if event.batch_id % 100 == 0:
with gzip.open("batch-" + str(event.batch_id) + ".tar.gz",
'w') as f:
trainer.save_parameter_to_tar(f)
result = trainer.test(
paddle.batch(
paddle.dataset.imikolov.test(word_dict, N), 32))
print "Pass %d, Batch %d, Cost %f, %s, Testing metrics %s" % (
event.pass_id, event.batch_id, event.cost, event.metrics,
result.metrics)
cost = paddle.layer.classification_cost(input=predictword, label=nextword)
parameters = paddle.parameters.create(cost)
adagrad = paddle.optimizer.AdaGrad(
learning_rate=3e-3,
regularization=paddle.optimizer.L2Regularization(8e-4))
trainer = paddle.trainer.SGD(cost,
parameters,
adagrad,
is_local=not cluster_train)
trainer.train(
paddle.batch(paddle.dataset.imikolov.train(word_dict, N), 32),
num_passes=30,
event_handler=event_handler)
if __name__ == '__main__':
main()
doc/howto/usage/cluster/src/word2vec/api_train_v2_cluster.py 0 → 100644
浏览文件 @ 06c7686a
import math
import os
import paddle.v2 as paddle
import pickle
embsize = 32
hiddensize = 256
N = 5
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")
def wordemb(inlayer):
wordemb = paddle.layer.embedding(
input=inlayer,
size=embsize,
param_attr=paddle.attr.Param(
name="_proj",
initial_std=0.001,
learning_rate=1,
l2_rate=0,
sparse_update=True))
return wordemb
def cluster_reader_cluster(filename, node_id):
def cluster_reader():
with open("-".join([filename, "%05d" % int(node_id)]), "r") as f:
for l in f:
csv_data = [int(cell) for cell in l.split(",")]
yield tuple(csv_data)
return cluster_reader
def main():
# get arguments from env
# for local training
TRUTH = ["true", "True", "TRUE", "1", "yes", "Yes", "YES"]
cluster_train = os.getenv('PADDLE_CLUSTER_TRAIN', "False") in TRUTH
use_gpu = os.getenv('PADDLE_INIT_USE_GPU', "False")
if not cluster_train:
paddle.init(
use_gpu=use_gpu,
trainer_count=int(os.getenv("PADDLE_INIT_TRAINER_COUNT", "1")))
else:
paddle.init(
use_gpu=use_gpu,
trainer_count=int(os.getenv("PADDLE_INIT_TRAINER_COUNT", "1")),
port=int(os.getenv("PADDLE_INIT_PORT", "7164")),
ports_num=int(os.getenv("PADDLE_INIT_PORTS_NUM", "1")),
ports_num_for_sparse=int(
os.getenv("PADDLE_INIT_PORTS_NUM_FOR_SPARSE", "1")),
num_gradient_servers=int(
os.getenv("PADDLE_INIT_NUM_GRADIENT_SERVERS", "1")),
trainer_id=int(os.getenv("PADDLE_INIT_TRAINER_ID", "0")),
pservers=os.getenv("PADDLE_INIT_PSERVERS", "127.0.0.1"))
fn = open("thirdparty/wuyi_train_thdpty/word_dict.pickle", "r")
word_dict = pickle.load(fn)
fn.close()
dict_size = len(word_dict)
firstword = paddle.layer.data(
name="firstw", type=paddle.data_type.integer_value(dict_size))
secondword = paddle.layer.data(
name="secondw", type=paddle.data_type.integer_value(dict_size))
thirdword = paddle.layer.data(
name="thirdw", type=paddle.data_type.integer_value(dict_size))
fourthword = paddle.layer.data(
name="fourthw", type=paddle.data_type.integer_value(dict_size))
nextword = paddle.layer.data(
name="fifthw", type=paddle.data_type.integer_value(dict_size))
Efirst = wordemb(firstword)
Esecond = wordemb(secondword)
Ethird = wordemb(thirdword)
Efourth = wordemb(fourthword)
contextemb = paddle.layer.concat(input=[Efirst, Esecond, Ethird, Efourth])
hidden1 = paddle.layer.fc(input=contextemb,
size=hiddensize,
act=paddle.activation.Sigmoid(),
layer_attr=paddle.attr.Extra(drop_rate=0.5),
bias_attr=paddle.attr.Param(learning_rate=2),
param_attr=paddle.attr.Param(
initial_std=1. / math.sqrt(embsize * 8),
learning_rate=1))
predictword = paddle.layer.fc(input=hidden1,
size=dict_size,
bias_attr=paddle.attr.Param(learning_rate=2),
act=paddle.activation.Softmax())
def event_handler(event):
if isinstance(event, paddle.event.EndIteration):
if event.batch_id % 100 == 0:
result = trainer.test(
paddle.batch(
cluster_reader_cluster(cluster_test_file, node_id), 32))
print "Pass %d, Batch %d, Cost %f, %s, Testing metrics %s" % (
event.pass_id, event.batch_id, event.cost, event.metrics,
result.metrics)
cost = paddle.layer.classification_cost(input=predictword, label=nextword)
parameters = paddle.parameters.create(cost)
adagrad = paddle.optimizer.AdaGrad(
learning_rate=3e-3,
regularization=paddle.optimizer.L2Regularization(8e-4))
trainer = paddle.trainer.SGD(cost,
parameters,
adagrad,
is_local=not cluster_train)
trainer.train(
paddle.batch(cluster_reader_cluster(cluster_train_file, node_id), 32),
num_passes=30,
event_handler=event_handler)
if __name__ == '__main__':
main()
doc/howto/usage/cluster/src/word2vec/prepare.py 0 → 100644
浏览文件 @ 06c7686a
import paddle.v2 as paddle
import tarfile
import os
import pickle
SPLIT_COUNT = 3
N = 5
def file_len(fd):
for i, l in enumerate(fd):
pass
return i + 1
def split_from_reader_by_line(filename, reader, split_count):
fn = open(filename, "w")
for batch_id, batch_data in enumerate(reader()):
batch_data_str = [str(d) for d in batch_data]
fn.write(",".join(batch_data_str))
fn.write("\n")
fn.close()
fn = open(filename, "r")
total_line_count = file_len(fn)
fn.close()
per_file_lines = total_line_count / split_count + 1
cmd = "split -d -a 5 -l %d %s %s-" % (per_file_lines, filename, filename)
os.system(cmd)
word_dict = paddle.dataset.imikolov.build_dict()
with open("word_dict.pickle", "w") as dict_f:
pickle.dump(word_dict, dict_f)
split_from_reader_by_line("train.txt",
paddle.dataset.imikolov.train(word_dict, N),
SPLIT_COUNT)
split_from_reader_by_line("test.txt",
paddle.dataset.imikolov.test(word_dict, N),
SPLIT_COUNT)
doc/howto/usage/cmd_parameter/arguments_cn.md
浏览文件 @ 06c7686a
......@@ -63,7 +63,7 @@
</tr>
<tr>
<td class="left" rowspan="15">训练</td><td class="left">dot_period</td>
<td class="left" rowspan="14">训练</td><td class="left">dot_period</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
......
doc/index_cn.rst
浏览文件 @ 06c7686a
......@@ -8,3 +8,4 @@ PaddlePaddle 文档
howto/index_cn.rst
api/index_cn.rst
faq/index_cn.rst
mobile/index_cn.rst
doc/index_en.rst
浏览文件 @ 06c7686a
......@@ -7,3 +7,4 @@ PaddlePaddle Documentation
getstarted/index_en.rst
howto/index_en.rst
api/index_en.rst
mobile/index_en.rst
doc/howto/cross_compiling/cross_compiling_for_android_cn.md → doc/mobile/cross_compiling_for_android_cn.md
浏览文件 @ 06c7686a
# 构建Android平台上的PaddlePaddle库
用户可通过交叉编译的方式，在用户熟悉的开发平台（Linux，Mac OS X和Windows）上编译Android平台上适用的PaddlePaddle库。
用户可通过如下两种方式，交叉编译Android平台上适用的PaddlePaddle库：
- 基于Docker容器的编译方式
- 基于Linux交叉编译环境的编译方式
## 基于Docker容器的编译方式
Docker能在所有主要操作系统（包括Linux，Mac OS X和Windows）上运行，因此，使用基于Docker容器的编译方式，用户可在自己熟悉的开发平台上编译Android平台上适用的PaddlePaddle库。
### 构建PaddlePaddle的Android开发镜像
我们把PaddlePaddle的交叉编译环境打包成一个镜像，称为开发镜像，里面涵盖了交叉编译Android版PaddlePaddle库需要的所有编译工具。
```bash
$ git clone https://github.com/PaddlePaddle/Paddle.git
$ cd Paddle
$ docker build -t username/paddle-android:dev . -f Dockerfile.android
```
### 编译PaddlePaddle C-API库
构建好开发镜像后，即可使用开发镜像来编译Android版PaddlePaddle C-API库。
Android的Docker开发镜像向用户提供两个可配置的参数：
<table class="docutils">
<colgroup>
<col width="25%" />
<col width="50%" />
<col width="25%" />
</colgroup>
<thead valign="bottom">
<tr class="row-odd">
<th class="head">Argument</th>
<th class="head">Optional Values</th>
<th class="head">Default</th>
</tr>
</thead>
<tbody valign="top">
<tr class="row-even">
<td>ANDROID_ABI</td>
<td>armeabi-v7a, arm64-v8a</td>
<td>armeabi-v7a</td>
</tr>
<tr class="row-odd">
<td>ANDROID_API</td>
<td>>= 21</td>
<td>21</td>
</tr>
</tbody>
</table>
- 编译`armeabi-v7a`，`Android API 21`的PaddlePaddle库
```bash
$ docker run -it --rm -v $PWD:/paddle -e "ANDROID_ABI=armeabi-v7a" -e "ANDROID_API=21" username/paddle-android:dev
```
- 编译`arm64-v8a`，`Android API 21`的PaddlePaddle库
```bash
$ docker run -it --rm -v $PWD:/paddle -e "ANDROID_ABI=arm64-v8a" -e "ANDROID_API=21" username/paddle-android:dev
```
执行上述`docker run`命令时，容器默认执行[paddle/scripts/docker/build_android.sh](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/scripts/docker/build_android.sh)脚本。该脚本中记录了交叉编译Android版PaddlePaddle库常用的CMake配置，并且会根据`ANDROID_ABI`和`ANDROID_API`自动构建独立工具链、进行编译和安装。由于arm64架构要求Android API不小于21。因此当`ANDROID_ABI=arm64-v8a`，`ANDROID_API<21`时，Docker容器中将默认使用`Android API 21`的编译工具链。用户可以参考下文**配置交叉编译参数**章节，根据个人的需求修改定制Docker容器所执行的脚本。编译安装结束之后，PaddlePaddle的C-API库将被安装到`$PWD/install_android`目录，所依赖的第三方库同时也被安装到`$PWD/install_android/third_party`目录。
## 基于Linux交叉编译环境的编译方式
本文档将以Linux x86-64平台为例，介绍交叉编译Android平台上适用的PaddlePaddle库的方法和步骤。
## 准备交叉编译环境
### 准备交叉编译环境
从源码交叉编译PaddlePaddle，用户需要提前准备好交叉编译环境。Android平台上使用的C/C++交叉编译工具链为[Android NDK](https://developer.android.com/ndk/downloads/index.html?hl=zh-cn)，用户可自行前往下载预编译好的版本，也可通过以下命令获取：
......@@ -13,18 +72,27 @@ unzip -q android-ndk-r14b-linux-x86_64.zip
```
Android NDK中包含了所有Android API级别、所有架构（arm/arm64/x86/mips）需要用到的编译工具和系统库。用户可根据自己的编译目标架构、所需支持的最低Android API级别，构建[独立工具链](https://developer.android.google.cn/ndk/guides/standalone_toolchain.html?hl=zh-cn)。
比如：
- 构建`armeabi-v7a`、 `Android API 21`的独立工具链：
```bash
your/path/to/android-ndk-r14b-linux-x86_64/build/tools/make-standalone-toolchain.sh \
--arch=arm --platform=android-21 --install-dir=your/path/to/my_standalone_toolchain
--arch=arm --platform=android-21 --install-dir=your/path/to/arm_standalone_toolchain
```
此命令将在your/path/to/my_standalone_toolchain目录生成一套编译工具链，面向架构为32位ARM架构，支持的最小的Android API级别为21，使用的编译器为arm-linux-androideabi-gcc (GCC) 4.9。
此命令将在`your/path/to/arm_standalone_toolchain`目录生成一套独立编译工具链，面向架构为32位ARM架构，支持的最小的Android API级别为21，支持编译器`arm-linux-androideabi-gcc (GCC) 4.9`和`clang 3.8`。
- 构建`arm64-v8a`、 `Android API 21`的独立工具链：
```bash
your/path/to/android-ndk-r14b-linux-x86_64/build/tools/make-standalone-toolchain.sh \
--arch=arm64 --platform=android-21 --install-dir=your/path/to/arm64_standalone_toolchain
```
注意：**PaddlePaddle要求使用的编译工具链所支持的Andoid API级别不小于21**。
此命令将在`your/path/to/arm64_standalone_toolchain`目录生成一套独立编译工具链，面向架构为64位ARM64架构，支持的最小Android API级别为21，支持编译器`arm-linux-androideabi-gcc (GCC) 4.9`和`clang 3.8`。
## 配置交叉编译参数
注意：**PaddlePaddle要求使用的编译工具链所支持的Android API级别不小于21**。
### 配置交叉编译参数
CMake系统对交叉编译提供了支持[cmake-toolchains](https://cmake.org/cmake/help/v3.0/manual/cmake-toolchains.7.html#cross-compiling)。为了简化cmake配置，PaddlePaddle为交叉编译提供了工具链配置文档[cmake/cross_compiling/android.cmake](https://github.com/PaddlePaddle/Paddle/blob/develop/cmake/cross_compiling/android.cmake)，以提供一些默认的编译器和编译参数相关配置。注意，从CMake 3.7版本开始，CMake官方对Android平台的交叉编译提供了通用的支持。PaddlePaddle若检测到用户使用的CMake版本不低于3.7时，将会将用户传进来的配置参数传递CMake系统，交由CMake系统本身来处理。有关参数配置的详细说明见[cmake-toolchains](https://cmake.org/cmake/help/v3.7/manual/cmake-toolchains.7.html#cross-compiling)。
......@@ -36,23 +104,43 @@ CMake系统对交叉编译提供了支持[cmake-toolchains](https://cmake.org/cm
Android平台可选配置参数：
- `ANDROID_STANDALONE_TOOLCHAIN`，独立工具链所在的绝对路径，或者相对于构建目录的相对路径。PaddlePaddle的CMake系统将根据该值自动推导和设置需要使用的交叉编译器、sysroot、以及Android API级别；否则，用户需要在cmake时手动设置这些值。无默认值。
- `ANDROID_ABI`，目标架构ABI。目前只支持`armeabi-v7a`，默认值为`armeabi-v7a`。
- `ANDROID_TOOLCHAIN`，目标工具链。可设置`gcc/clang`，默认值为`clang`。
- CMake 3.7以上，将会始终使用`clang`工具链；CMake 3.7以下，可设置`ANDROID_TOOLCHAIN=gcc`以使用`gcc`工具链。
- Android官方提供的`clang`编译器要求系统支持`GLIBC 2.15`以上。
- `ANDROID_ABI`，目标架构ABI。目前支持`armeabi-v7a`和`arm64-v8a`，默认值为`armeabi-v7a`。
- `ANDROID_NATIVE_API_LEVEL`，工具链的Android API级别。若没有显式设置，PaddlePaddle将根据`ANDROID_STANDALONE_TOOLCHAIN`的值自动推导得到。
- `ANROID_ARM_MODE`，是否使用ARM模式。可设置`ON/OFF`，默认值为`ON`。
- `ANDROID_ARM_NEON`，是否使用NEON指令。目前必须设置成`ON`，默认值为`ON`。
- `ANROID_ARM_MODE`，是否使用ARM模式。
- `ANDROID_ABI=armeabi-v7a`时，可设置`ON/OFF`，默认值为`ON`；
- `ANDROID_ABI=arm64-v8a`时，不需要设置。
- `ANDROID_ARM_NEON`，是否使用NEON指令。
- `ANDROID_ABI=armeabi-v7a`时，可设置`ON/OFF`，默认值为`ON`；
- `ANDROID_ABI=arm64-v8a`时，不需要设置。
其他配置参数：
- `USE_EIGEN_FOR_BLAS`，是否使用Eigen库进行矩阵计算。可设置`ON/OFF`，默认值为`OFF`。
- `HOST_C/CXX_COMPILER`，宿主机的C/C++编译器。在编译宿主机版protoc可执行文件和目标机版OpenBLAS库时需要用到。默认设置成环境变量`CC`的值；若环境变量`CC`没有设置，则设置成`cc`编译器。
一种常用的cmake配置如下：
常用的cmake配置如下：
```bash
cmake -DCMAKE_SYSTEM_NAME=Android \
-DANDROID_STANDALONE_TOOLCHAIN=your/path/to/my_standalone_toolchain \
-DANDROID_STANDALONE_TOOLCHAIN=your/path/to/arm_standalone_toolchain \
-DANDROID_ABI=armeabi-v7a \
-DANDROID_ARM_NEON=ON \
-DANDROID_ARM_MODE=ON \
-DUSE_EIGEN_FOR_BLAS=ON \
-DCMAKE_INSTALL_PREFIX=your/path/to/install \
-DWITH_C_API=ON \
-DWITH_SWIG_PY=OFF \
..
```
```
cmake -DCMAKE_SYSTEM_NAME=Android \
-DANDROID_STANDALONE_TOOLCHAIN=your/path/to/arm64_standalone_toolchain \
-DANDROID_ABI=arm64-v8a \
-DUSE_EIGEN_FOR_BLAS=OFF \
-DCMAKE_INSTALL_PREFIX=your/path/to/install \
-DWITH_C_API=ON \
-DWITH_SWIG_PY=OFF \
......@@ -61,7 +149,12 @@ cmake -DCMAKE_SYSTEM_NAME=Android \
用户还可根据自己的需求设置其他编译参数。比如希望最小化生成的库的大小，可以设置`CMAKE_BUILD_TYPE`为`MinSizeRel`；若希望最快的执行速度，则可设置`CMAKE_BUILD_TYPE`为`Release`。亦可以通过手动设置`CMAKE_C/CXX_FLAGS_MINSIZEREL/RELEASE`来影响PaddlePaddle的编译过程。
## 编译和安装
**性能TIPS**，为了达到最快的计算速度，在CMake参数配置上，有以下建议：
- 设置`CMAKE_BUILD_TYPE`为`Release`
- 使用`clang`编译工具链
- `armeabi-v7a`时，设置`USE_EIGEN_BLAS=ON`，使用Eigen进行矩阵计算；`arm64-v8a`时，设置`USE_EIGEN_FOR_BLAS=OFF`，使用OpenBLAS进行矩阵计算
### 编译和安装
CMake配置完成后，执行以下命令，PaddlePaddle将自动下载和编译所有第三方依赖库、编译和安装PaddlePaddle预测库。
......@@ -72,4 +165,4 @@ make install
注意：如果你曾经在源码目录下编译过其他平台的PaddlePaddle库，请先使用`rm -rf`命令删除`third_party`目录和`build`目录，以确保所有的第三方依赖库和PaddlePaddle代码都是针对新的CMake配置重新编译的。
执行完安装命令后，`your/path/to/install`目录中会包含`include`和`lib`目录，其中`include`中包含C-API的头文件，`lib`中包含一个Android版本的库。自此，PaddlePaddle的已经安装完成，用户可将`your/path/to/install`目录下的生成文件用于深度学习相关Android App中，调用方法见C-API文档。
执行完安装命令后，`your/path/to/install`目录中会包含`include`、`lib`和`third_party`目录，其中`include`中包含C-API的头文件，`lib`中包含若干个不同Android ABI的PaddlePaddle库，`third_party`中包含所依赖的所有第三方库。自此，PaddlePaddle的已经安装完成，用户可将`your/path/to/install`目录下的生成文件用于深度学习相关Android App中，调用方法见C-API文档。
doc/mobile/cross_compiling_for_android_en.md 0 → 100644
浏览文件 @ 06c7686a
# Build PaddlePaddle for Android
There are two approaches to build PaddlePaddle for Android: using Docker and on Linux without Docker.
## Cross-Compiling Using Docker
Docker-based cross-compiling is the recommended approach because Docker runs on all major operating systems, including Linux, Mac OS X, and Windows.
### Build the Docker Image
The following steps pack all the tools that we need to build PaddlePaddle into a Docker image.
```bash
$ git clone https://github.com/PaddlePaddle/Paddle.git
$ cd Paddle
$ docker build -t paddle:dev-android . -f Dockerfile.android
```
### Build the Inference Library
We can run the Docker image we just created to build the inference library of PaddlePaddle for Android using the command below:
```bash
$ docker run -it --rm -v $PWD:/paddle -e "ANDROID_ABI=armeabi-v7a" -e "ANDROID_API=21" paddle:dev-android
```
The Docker image accepts two arguments `ANDROID_ABI` and `ANDROID_API`:
<table class="docutils">
<colgroup>
<col width="25%" />
<col width="50%" />
<col width="25%" />
</colgroup>
<thead valign="bottom">
<tr class="row-odd">
<th class="head">Argument</th>
<th class="head">Optional Values</th>
<th class="head">Default</th>
</tr>
</thead>
<tbody valign="top">
<tr class="row-even">
<td>ANDROID_ABI</td>
<td>armeabi-v7a, arm64-v8a</td>
<td>armeabi-v7a</td>
</tr>
<tr class="row-odd">
<td>ANDROID_API</td>
<td>>= 21</td>
<td>21</td>
</tr>
</tbody>
</table>
The ARM-64 architecture (`arm64-v8a`) requires at least level 21 of Android API.
The default entry-point of the Docker image, [`paddle/scripts/docker/build_android.sh`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/scripts/docker/build_android.sh) generates the [Android cross-compiling standalone toolchain](https://developer.android.com/ndk/guides/standalone_toolchain.html) based on the argument: `ANDROID_ABI` or `ANDROID_API`. For information about other configuration arguments, please continue reading.
The above command generates and outputs the inference library in `$PWD/install_android` and puts third-party libraries in `$PWD/install_android/third_party`.
## Cross-Compiling on Linux
The Linux-base approach to cross-compile is to run steps in `Dockerfile.android` manually on a Linux x64 computer.
### Setup the Environment
To build for Android's, we need [Android NDK](
https://developer.android.com/ndk/downloads/index.html):
```bash
wget -q https://dl.google.com/android/repository/android-ndk-r14b-linux-x86_64.zip
unzip -q android-ndk-r14b-linux-x86_64.zip
```
Android NDK includes everything we need to build the [*standalone toolchain*](https://developer.android.com/ndk/guides/standalone_toolchain.html), which in then used to build PaddlePaddle for Android. (We plan to remove the intermediate stage of building the standalone toolchain in the near future.)
- To build the standalone toolchain for `armeabi-v7a` and Android API level 21:
```bash
your/path/to/android-ndk-r14b-linux-x86_64/build/tools/make-standalone-toolchain.sh \
--arch=arm --platform=android-21 --install-dir=your/path/to/arm_standalone_toolchain
```
The generated standalone toolchain will be in `your/path/to/arm_standalone_toolchain`.
- To build the standalone toolchain for `arm64-v8a` and Android API level 21:
```bash
your/path/to/android-ndk-r14b-linux-x86_64/build/tools/make-standalone-toolchain.sh \
--arch=arm64 --platform=android-21 --install-dir=your/path/to/arm64_standalone_toolchain
```
The generated standalone toolchain will be in `your/path/to/arm64_standalone_toolchain`.
**Please be aware that the minimum level of Android API required by PaddlePaddle is 21.**
### Cross-Compiling Arguments
CMake supports [choosing the toolchain](https://cmake.org/cmake/help/v3.0/manual/cmake-toolchains.7.html#cross-compiling). PaddlePaddle provides [`android.cmake`](https://github.com/PaddlePaddle/Paddle/blob/develop/cmake/cross_compiling/android.cmake), which configures the Android cross-compiling toolchain for CMake. `android.cmake` is not required for CMake >= 3.7, which support Android cross-compiling. PaddlePaddle detects the CMake version, for those newer than 3.7, it uses [the official version](https://cmake.org/cmake/help/v3.7/manual/cmake-toolchains.7.html#cross-compiling).
Some other CMake arguments you need to know:
- `CMAKE_SYSTEM_NAME` must be `Android`. This tells PaddlePaddle's CMake system to cross-compile third-party dependencies. This also changes some other CMake arguments like `WITH_GPU=OFF`, `WITH_AVX=OFF`, `WITH_PYTHON=OFF`, and `WITH_RDMA=OFF`.
- `WITH_C_API` must be `ON`, to build the C-based inference library for Android.
- `WITH_SWIG_PY` must be `OFF` because the Android platform doesn't support SWIG-based API.
Some Android-specific arguments:
- `ANDROID_STANDALONE_TOOLCHAIN`: the absolute path of the Android standalone toolchain, or the path relative to the CMake build directory. PaddlePaddle's CMake extensions would derive the cross-compiler, sysroot and Android API level from this argument.
- `ANDROID_TOOLCHAIN`: could be `gcc` or `clang`. The default value is `clang`.
- For CMake >= 3.7, it should anyway be `clang`. For older versions, it could be `gcc`.
- Android's official `clang` requires `glibc` >= 2.15.
- `ANDROID_ABI`: could be `armeabi-v7a` or `arm64-v8a`. The default value is `armeabi-v7a`.
- `ANDROID_NATIVE_API_LEVEL`: could be derived from the value of `ANDROID_STANDALONE_TOOLCHAIN`.
- `ANROID_ARM_MODE`:
- could be `ON` or `OFF`, and defaults to `ON`, when `ANDROID_ABI=armeabi-v7a`;
- no need to specify when `ANDROID_ABI=arm64-v8a`.
- `ANDROID_ARM_NEON`: indicates if to use NEON instructions.
- could be `ON` or `OFF`, and defaults to `ON`, when `ANDROID_ABI=armeabi-v7a`;
- no need to specify when `ANDROID_ABI=arm64-v8a`.
Other useful arguments:
- `USE_EIGEN_FOR_BLAS`: indicates if using Eigen. Could be `ON` or `OFF`, defaults to `OFF`.
- `HOST_C/CXX_COMPILER`: specifies the host compiler, which is used to build the host-specific protoc and target-specific OpenBLAS. It defaults to the value of the environment variable `CC`, or `cc`.
Some frequent configurations for your reference:
```bash
cmake -DCMAKE_SYSTEM_NAME=Android \
-DANDROID_STANDALONE_TOOLCHAIN=your/path/to/arm_standalone_toolchain \
-DANDROID_ABI=armeabi-v7a \
-DANDROID_ARM_NEON=ON \
-DANDROID_ARM_MODE=ON \
-DUSE_EIGEN_FOR_BLAS=ON \
-DCMAKE_INSTALL_PREFIX=your/path/to/install \
-DWITH_C_API=ON \
-DWITH_SWIG_PY=OFF \
..
```
```
cmake -DCMAKE_SYSTEM_NAME=Android \
-DANDROID_STANDALONE_TOOLCHAIN=your/path/to/arm64_standalone_toolchain \
-DANDROID_ABI=arm64-v8a \
-DUSE_EIGEN_FOR_BLAS=OFF \
-DCMAKE_INSTALL_PREFIX=your/path/to/install \
-DWITH_C_API=ON \
-DWITH_SWIG_PY=OFF \
..
```
There are some other arguments you might want to configure.
- `CMAKE_BUILD_TYPE=MinSizeRel` minimizes the size of library.
- `CMAKE_BUILD_TYPE-Release` optimizes the runtime performance.
Our own tip for performance optimization to use clang and Eigen or OpenBLAS:
- `CMAKE_BUILD_TYPE=Release`
- `ANDROID_TOOLCHAIN=clang`
- `USE_EIGEN_BLAS=ON` for `armeabi-v7a`, or `USE_EIGEN_FOR_BLAS=OFF` for `arm64-v8a`.
### Build and Install
After running `cmake`, we can run `make; make install` to build and install.
Before building, you might want to remove the `third_party` and `build` directories including pre-built libraries for other architectures.
After building，in the directory `CMAKE_INSTALL_PREFIX`, you will find three sub-directories:
- `include`: the header file of the inference library,
- `lib`: the inference library built for various Android ABIs,
- `third_party`: dependent third-party libraries built for Android.
doc/mobile/cross_compiling_for_ios_cn.md 0 → 100644
浏览文件 @ 06c7686a
# 构建iOS平台上的PaddlePaddle库
交叉编译iOS平台上适用的PaddlePaddle库，需要在MacOS系统上进行。本文的将介绍在MacOS上，从源码交叉编译iOS平台上适用的PaddlePaddle库。
## 准备交叉编译环境
Apple官方为iOS开发提供了完整的交叉编译工具和集成开发环境，用户从App Store下载安装Xcode即可。也可自行前往官网下载，[Xcode](https://developer.apple.com/cn/xcode/)。安装完成之后，可在命令行执行`xcodebuild -version`，判断是否安装成功。
```bash
$ xcodebuild -version
Xcode 9.0
Build version 9A235
```
## 配置交叉编译参数
PaddlePaddle为交叉编译提供了工具链配置文档[cmake/cross_compiling/ios.cmake](https://github.com/PaddlePaddle/Paddle/blob/develop/cmake/cross_compiling/ios.cmake)，以提供一些默认的编译器和编译参数配置。
交叉编译iOS版本的PaddlePaddle库时，有一些必须配置的参数：
- `CMAKE_SYSTEM_NAME`，CMake编译的目标平台，必须设置为`iOS`。在设置`CMAKE_SYSTEM_NAME=iOS`后，PaddlePaddle的CMake系统会自动编译所有的第三方依赖库，并且强制设置一些PaddlePaddle参数的值（`WITH_C_API=ON`、`WITH_GPU=OFF`、`WITH_AVX=OFF`、`WITH_PYTHON=OFF`、`WITH_RDMA=OFF`）。
- `WITH_C_API`，是否编译C-API预测库，必须设置为ON。在iOS平台上只支持使用C-API来预测。
- `WITH_SWIG_PY`，必须设置为ON。在iOS平台上不支持通过swig调用来训练或者预测。
iOS平台可选配置参数：
- `IOS_PLATFORM`，可设置为`OS/SIMULATOR`，默认值为`OS`。
- `OS`，构建目标为`arm`架构的iPhone或者iPad等物理设备。
- `SIMULATOR`，构建目标为`x86`架构的模拟器平台。
- `IOS_ARCH`，目标架构。针对不同的`IOS_PLATFORM`，可设置的目标架构如下表所示：
<table class="docutils">
<colgroup>
<col width="35%" />
<col width="65%" />
</colgroup>
<thead valign="bottom">
<tr class="row-odd">
<th class="head">IOS_PLATFORM</th>
<th class="head">IOS_ARCH</th>
</tr>
</thead>
<tbody valign="top">
<tr class="row-even">
<td>OS</td>
<td>armv7, armv7s, arm64 (默认)</td>
</tr>
<tr class="row-odd">
<td>SIMULATOR</td>
<td>i386, x86_64 (默认)</td>
</tr>
</tbody>
</table>
- `IOS_DEPLOYMENT_TARGET`，最小的iOS部署版本，默认值为`7.0`。
- `IOS_ENABLE_BITCODE`，是否使能[Bitcode](https://developer.apple.com/library/content/documentation/IDEs/Conceptual/AppDistributionGuide/AppThinning/AppThinning.html#//apple_ref/doc/uid/TP40012582-CH35-SW3)，可设置`ON/OFF`，默认值为`ON`。
- `IOS_USE_VECLIB_FOR_BLAS`，是否使用[vecLib](https://developer.apple.com/documentation/accelerate/veclib)框架进行BLAS矩阵计算，可设置`ON/OFF`，默认值为`OFF`。
- `IOS_DEVELOPMENT_ROOT`，`Developer`目录，可显式指定为`/path/to/platform/Developer`。若未显式指定，PaddlePaddle将会根据`IOS_PLATFORM`自动选择`Xcode`对应`platform`的`Developer`目录。
- `IOS_SDK_ROOT`，所使用`SDK`的根目录，可显式指定为`/path/to/platform/Developer/SDKs/SDK`。若未显式指定，PaddlePaddle将会自动选择`IOS_DEVELOPMENT_ROOT`目录下最新的`SDK`版本。
其他配置参数：
- `USE_EIGEN_FOR_BLAS`，是否使用Eigen库进行矩阵计算，在`IOS_USE_VECLIB_FOR_BLAS=OFF`时有效。可设置`ON/OFF`，默认值为`OFF`。
- `HOST_C/CXX_COMPILER`，宿主机的C/C++编译器。默认值为环境变量`CC/CXX`的值；若环境变量`CC/CXX`未设置，则使用`cc/c++`编译器。
常用的cmake配置如下：
```bash
cmake -DCMAKE_SYSTEM_NAME=iOS \
-DIOS_PLATFORM=OS \
-DIOS_ARCH="arm64" \
-DIOS_ENABLE_BITCODE=ON \
-DIOS_USE_VECLIB_FOR_BLAS=ON \
-DCMAKE_INSTALL_PREFIX=your/path/to/install \
-DWITH_C_API=ON \
-DWITH_TESTING=OFF \
-DWITH_SWIG_PY=OFF \
..
```
```bash
cmake -DCMAKE_SYSTEM_NAME=iOS \
-DIOS_PLATFORM=SIMULATOR \
-DIOS_ARCH="x86_64" \
-DIOS_USE_VECLIB_FOR_BLAS=ON \
-DCMAKE_INSTALL_PREFIX=your/path/to/install \
-DWITH_C_API=ON \
-DWITH_TESTING=OFF \
-DWITH_SWIG_PY=OFF \
..
```
用户还可根据自己的需求设置其他编译参数。比如希望最小化生成库的大小，可以设置`CMAKE_BUILD_TYPE`为`MinSizeRel`；若希望得到最快的执行速度，则可设置`CMAKE_BUILD_TYPE`为`Release`。亦可以通过手动设置`CMAKE_C/CXX_FLAGS`来影响PaddlePaddle的编译过程。
**性能TIPS**，为了达到最快的计算速度，在CMake参数配置上，有以下建议：
- 设置`CMAKE_BUILD_TYPE`为`Release`
- 设置`IOS_USE_VECLIB_FOR_BLAS=ON`，调用`vecLib`框架提供的BLAS函数进行矩阵计算。
## 编译和安装
CMake配置完成后，执行以下命令，PaddlePaddle将自动下载和编译所有第三方依赖库、编译和安装PaddlePaddle预测库。
```
$ make
$ make install
```
注意：如果你曾在源码目录下编译过其他平台的PaddlePaddle库，请先使用`rm -rf`命令删除`third_party`目录和`build`目录，以确保所有的第三方依赖库和PaddlePaddle代码都是针对新的CMake配置重新编译的。
执行完安装命令后，`your/path/to/install`目录中会包含以下内容：
- `include`目录，其中包含所有C-API的头文件
- `lib`目录，其中包含PaddlePaddle的C-API静态库
- `third_party`目录，其中包含所依赖的所有第三方库
注意，不同架构的PaddlePaddle库建议安装到不同的目录下，然后使用`lipo`工具将多个静态库合并成一个支持多个架构的fat库。
自此，PaddlePaddle库已经安装完成，用户可将合成的fat库用于深度学习相关的iOS App中，调用方法见C-API文档。
doc/howto/cross_compiling/cross_compiling_for_raspberry_cn.md → doc/mobile/cross_compiling_for_raspberry_cn.md
浏览文件 @ 06c7686a
# 构建Raspberry Pi平台上的PaddlePaddle库
对于Rasspberry Pi系统，用户可通过ssh等方式登录到Raspberry Pi系统上，按照[源码编译PaddlePaddle](http://www.paddlepaddle.org/doc_cn/getstarted/build_and_install/cmake/build_from_source_cn.html)相关文档所述，直接编译Raspberry Pi平台上适用的PaddlePaddle库。
通常有两个方法来构建基于 Rasspberry Pi 的版本：
用户也可以在自己熟悉的开发平台上，通过交叉编译的方式来编译。这篇文档将以Linux x86-64平台为例，介绍交叉编译Raspberry Pi平台上适用的PaddlePaddle的方法和步骤。
1. 通过ssh等方式登录到Raspberry Pi系统上来构建。所需的开发工具和第三方库可以参考 [`/Dockerfile`](https://github.com/PaddlePaddle/Paddle/blob/develop/Dockerfile)。
## 准备交叉编译环境
1. 另一个方法是交叉编译。这篇文档介绍在 Linux/x64 上交叉编译Raspberry Pi平台上适用的PaddlePaddle的方法和步骤。
从源码交叉编译PaddlePaddle，用户需要提前准备好交叉编译环境。用户可自行前往[github](https://github.com/raspberrypi/tools)下载Raspberry Pi平台使用的C/C++交叉编译工具链，也可通过以下命令获取：
## 安装交叉编译器
克隆下面 Github repo
```bash
git clone https://github.com/raspberrypi/tools.git
```
该github仓库中包含若干个预编译好的、针对不同平台的编译工具。宿主机是Linux x86-64环境，则需选用`arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian-x64`下的作为编译工具，所使用的编译器为arm-linux-gnueabihf-gcc 4.8.3。
注意，该编译工具链需要系统glibc支持2.14以上。
即可在 `./tools/tree/master/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian-x64` 目录里找到交叉编译器 arm-linux-gnueabihf-gcc 4.8.3。运行该编译工具链需要一台 Linux x64 机器上以及 2.14版本以上的 glibc。
## 配置交叉编译参数
CMake系统对交叉编译提供了支持[cmake-toolchains](https://cmake.org/cmake/help/v3.0/manual/cmake-toolchains.7.html#cross-compiling)。为了简化cmake配置，PaddlePaddle为交叉编译提供了工具链配置文档[cmake/cross_compiling/raspberry_pi.cmake](https://github.com/PaddlePaddle/Paddle/blob/develop/cmake/cross_compiling/raspberry_pi.cmake)，以提供一些默认的编译器和编译参数相关配置。
CMake[支持交叉编译](https://cmake.org/cmake/help/v3.0/manual/cmake-toolchains.7.html#cross-compiling)。PaddlePaddle for Raspberry Pi的配置信息在[cmake/cross_compiling/raspberry_pi.cmake](https://github.com/PaddlePaddle/Paddle/blob/develop/cmake/cross_compiling/raspberry_pi.cmake)。
交叉编译Raspberry Pi版本PaddlePaddle库时，有一些必须配置的参数：
- `CMAKE_SYSTEM_NAME`，CMake编译的目标平台，必须配置为`RPi`。在设置`CMAKE_SYSTEM_NAME=RPi`后，PaddlePaddle的CMake系统才认为在是在交叉编译Raspberry Pi系统的版本，并自动编译宿主机版protoc可执行文件、目标机版protobuf库、以及目标机版OpenBLAS库。
Raspberry Pi平台可选配置参数：
- `CMAKE_SYSTEM_NAME`：CMake编译的目标平台，必须配置为`RPi`。在设置`CMAKE_SYSTEM_NAME=RPi`后，PaddlePaddle的CMake系统才认为在是在交叉编译Raspberry Pi系统的版本，并自动编译宿主机版protoc可执行文件、目标机版protobuf库、以及目标机版OpenBLAS库。
- `RPI_TOOLCHAIN`，编译工具链所在的绝对路径，或者相对于构建目录的相对路径。PaddlePaddle的CMake系统将根据该值自动设置需要使用的交叉编译器；否则，用户需要在cmake时手动设置这些值。无默认值。
- `RPI_ARM_NEON`，是否使用NEON指令。目前必须设置成`ON`，默认值为`ON`。
- `RPI_TOOLCHAIN`：编译工具链所在的绝对路径，或者相对于构建目录的相对路径。PaddlePaddle的CMake系统将根据该值自动设置需要使用的交叉编译器；否则，用户需要在cmake时手动设置这些值。无默认值。
其他配置参数：
- `RPI_ARM_NEON`：是否使用NEON指令。目前必须设置成`ON`，默认值为`ON`。
- `HOST_C/CXX_COMPILER`，宿主机的C/C++编译器。在编译宿主机版protoc可执行文件和目标机版OpenBLAS库时需要用到。默认设置成环境变量`CC`的值；若环境变量`CC`没有设置，则设置成`cc`编译器。
cmake参数如下；
一个常用的CMake配置如下：
```
cmake -DCMAKE_SYSTEM_NAME=RPi \
......@@ -47,7 +44,9 @@ cmake -DCMAKE_SYSTEM_NAME=RPi \
..
```
用户还可根据自己的需求设置其他编译参数。比如希望最小化生成的库的大小，可以设置`CMAKE_BUILD_TYPE`为`MinSizeRel`；若希望最快的执行速度，则可设置`CMAKE_BUILD_TYPE`为`Release`。亦可以通过手动设置`CMAKE_C/CXX_FLAGS_MINSIZEREL/RELEASE`来影响PaddlePaddle的编译过程。
其中`WITH_C_API=ON`表示需要构建推理库。
用户还可根据自己的需求设置其他编译参数。比如希望最小化生成的库的大小，可以设置`CMAKE_BUILD_TYPE`为`MinSizeRel`；若希望最快的执行速度，则可设置`CMAKE_BUILD_TYPE`为`Release`。
## 编译和安装
......@@ -60,6 +59,4 @@ make install
注意：如果你曾经在源码目录下编译过其他平台的PaddlePaddle库，请先使用`rm -rf`命令删除`third_party`目录和`build`目录，以确保所有的第三方依赖库和PaddlePaddle代码都是针对新的CMake配置重新编译的。
执行完安装命令后，由于上一步cmake配置中`WITH_C_API`设置为`ON`，`your/path/to/install`目录中会包含`include`和`lib`目录，其中`include`中包含C-API的头文件，`lib`中包含一个Raspberry Pi版本的库。
更多的编译配置见[源码编译PaddlePaddle](http://www.paddlepaddle.org/doc_cn/getstarted/build_and_install/cmake/build_from_source_cn.html)相关文档。
执行完安装命令后，`your/path/to/install`目录中会包含`include`和`lib`目录，其中`include`中包含C-API的头文件，`lib`中包含一个Raspberry Pi版本的库。
doc/mobile/cross_compiling_for_raspberry_en.md 0 → 100644
浏览文件 @ 06c7686a
# Build PaddlePaddle for Raspberry Pi
You may use any of the following two approaches to build the inference library of PaddlePaddle for Raspberry Pi:
1. Build using SSH: Log in to a Raspberry Pi using SSH and build the library. The required development tools and third-party dependencies are listed in here: [`/Dockerfile`](https://github.com/PaddlePaddle/Paddle/blob/develop/Dockerfile).
1. Cross-compile: We talk about how to cross-compile PaddlePaddle for Raspberry Pi on a Linux/x64 machine, in more detail in this article.
## The Cross-Compiling Toolchain
Step 1. Clone the Github repo by running the following command.
```bash
git clone https://github.com/raspberrypi/tools.git
```
Step 2. Use the pre-built cross-compiler found in `./tools/tree/master/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian-x64`. To run it on a Linux computer, glibc version >= 2.14 is needed.
## CMake Arguments
CMake supports [cross-compiling](https://cmake.org/cmake/help/v3.0/manual/cmake-toolchains.7.html#cross-compiling). All CMake configuration arguments required for the cross-compilation for Raspberry Pi can be found in [`cmake/cross_compiling/raspberry_pi.cmake`](https://github.com/PaddlePaddle/Paddle/blob/develop/cmake/cross_compiling/raspberry_pi.cmake).
Some important arguments that need to be set:
- `CMAKE_SYSTEM_NAME`: The target platform. Must be `RPi`.
- `RPI_TOOLCHAIN`: The absolute path of the cross-compiling toolchain.
- `RPI_ARM_NEON`: Use ARM NEON Intrinsics. This is a required argument and set default to `ON`.
- `HOST_C/CXX_COMPILER`: The C/C++ compiler for the host. It is used to build building tools running on the host, for example, protoc.
A commonly-used CMake configuration is as follows:
```
cmake -DCMAKE_SYSTEM_NAME=RPi \
-DRPI_TOOLCHAIN=your/path/to/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian-x64 \
-DRPI_ARM_NEON=ON \
-DCMAKE_INSTALL_PREFIX=your/path/to/install \
-DWITH_GPU=OFF \
-DWITH_C_API=ON \
-DWITH_PYTHON=OFF \
-DWITH_SWIG_PY=OFF \
..
```
To build the inference library, please set the argument WITH\_C\_API to ON: `WITH_C_API=ON`.
You can add more arguments. For example, to minimize the size of the generated inference library, you may use `CMAKE_BUILD_TYPE=MinSizeRel`. For performance optimization, you may use `CMAKE_BUILD_TYPE=Release`.
## Build and Install
The following commands build the inference library of PaddlePaddle for Raspberry Pi and third-party dependencies.
```bash
make
make install
```
The intermediate files will be stored in `build`. Third-party libraries will be located in `build/third_party`. If you have already built it for other platforms like Android or iOS, you may want to clear these directories by running the command: `rm -rf build`.
The infernece library will be in `your/path/to/install/lib`, with related header files in `your/path/to/install/include`.
doc/mobile/index_cn.rst 0 → 100644
浏览文件 @ 06c7686a
MOBILE
======
.. toctree::
:maxdepth: 1
cross_compiling_for_android_cn.md
cross_compiling_for_ios_cn.md
cross_compiling_for_raspberry_cn.md
doc/mobile/index_en.rst 0 → 100644
浏览文件 @ 06c7686a
MOBILE
======
.. toctree::
:maxdepth: 1
cross_compiling_for_android_en.md
cross_compiling_for_raspberry_en.md
go/.gitignore
浏览文件 @ 06c7686a
vendor/
.glide/
proto/*.go
go/cmd/master/master.go
浏览文件 @ 06c7686a
......@@ -25,9 +25,8 @@ import (
"strings"
"time"
log "github.com/inconshreveable/log15"
"github.com/namsral/flag"
log "github.com/sirupsen/logrus"
"github.com/topicai/candy"
"github.com/PaddlePaddle/Paddle/go/master"
"github.com/PaddlePaddle/Paddle/go/utils/networkhelper"
......@@ -41,16 +40,20 @@ func main() {
taskTimeoutMax := flag.Int("task-timeout-max", 3, "max timtout count for each task before it being declared failed task.")
chunkPerTask := flag.Int("chunk-per-task", 10, "chunk per task.")
logLevel := flag.String("log-level", "info",
"log level, possible values: debug, info, warning, error, fatal, panic")
"log level, possible values: debug, info, warn, error, crit")
flag.Parse()
level, e := log.ParseLevel(*logLevel)
candy.Must(e)
lvl, err := log.LvlFromString(*logLevel)
if err != nil {
panic(err)
}
log.SetLevel(level)
log.Root().SetHandler(
log.LvlFilterHandler(lvl, log.CallerStackHandler("%+v", log.StderrHandler)),
)
if *endpoints == "" {
log.Warningln("-endpoints not set, fault tolerance not be enabled.")
log.Warn("-endpoints not set, fault tolerance not be enabled.")
}
var store master.Store
......@@ -58,23 +61,25 @@ func main() {
eps := strings.Split(*endpoints, ",")
ip, err := networkhelper.GetExternalIP()
if err != nil {
log.Fatal(err)
log.Crit("get external ip error", log.Ctx{"error": err})
panic(err)
}
addr := fmt.Sprintf("%s:%d", ip, *port)
store, err = master.NewEtcdClient(eps, addr, master.DefaultLockPath, master.DefaultAddrPath, master.DefaultStatePath, *ttlSec)
if err != nil {
log.Fatal(err)
log.Crit("error creating etcd client.", log.Ctx{"error": err})
panic(err)
}
} else {
store = &master.InMemStore{}
}
shutdown := func() {
log.Infoln("shutting down gracefully")
log.Info("shutting down gracefully")
err := store.Shutdown()
if err != nil {
log.Errorln(err)
log.Error("shutdown error", log.Ctx{"error": err})
}
}
......@@ -86,24 +91,28 @@ func main() {
s, err := master.NewService(store, *chunkPerTask, *taskTimeoutDur, *taskTimeoutMax)
if err != nil {
log.Fatal(err)
log.Crit("error creating new service.", log.Ctx{"error": err})
panic(err)
}
err = rpc.Register(s)
if err != nil {
log.Fatal(err)
log.Crit("error registering to etcd.", log.Ctx{"error": err})
panic(err)
}
rpc.HandleHTTP()
l, err := net.Listen("tcp", ":"+strconv.Itoa(*port))
if err != nil {
log.Fatal(err)
log.Crit("error listing to port", log.Ctx{"error": err, "port": *port})
panic(err)
}
go func() {
err = http.Serve(l, nil)
if err != nil {
log.Fatal(err)
log.Crit("error serving HTTP", log.Ctx{"error": err})
panic(err)
}
}()
......
go/cmd/pserver/pserver.go
浏览文件 @ 06c7686a
......@@ -27,11 +27,11 @@ import (
"github.com/topicai/candy"
"github.com/PaddlePaddle/Paddle/go/pserver"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
)
func main() {
port := flag.Int("port", 0, "port of the pserver")
port := flag.Int("port", 8001, "port of the pserver")
index := flag.Int("index", -1, "index of the pserver, set to -1 if use etcd for auto pserver index registry")
etcdEndpoint := flag.String("etcd-endpoint", "http://127.0.0.1:2379",
"comma separated endpoint string for pserver to connect to etcd")
......@@ -41,13 +41,17 @@ func main() {
checkpointPath := flag.String("checkpoint-path", "/checkpoints/", "save checkpoint path")
checkpointInterval := flag.Duration("checkpoint-interval", 600*time.Second, "save checkpoint per interval seconds")
logLevel := flag.String("log-level", "info",
"log level, possible values: debug, info, warning, error, fatal, panic")
"log level, possible values: debug, info, warn, error, crit")
flag.Parse()
level, err := log.ParseLevel(*logLevel)
candy.Must(err)
lvl, err := log.LvlFromString(*logLevel)
if err != nil {
panic(err)
}
log.SetLevel(level)
log.Root().SetHandler(
log.LvlFilterHandler(lvl, log.CallerStackHandler("%+v", log.StderrHandler)),
)
var idx int
......@@ -63,7 +67,7 @@ func main() {
cp, err = pserver.LoadCheckpoint(e, idx)
if err != nil {
if err == pserver.ErrCheckpointNotFound {
log.Infof("Could not find the pserver checkpoint.")
log.Info("load checkpoint error", "error", err)
} else {
panic(err)
}
......@@ -71,10 +75,10 @@ func main() {
}
shutdown := func() {
log.Infoln("shutting down gracefully")
log.Info("shutting down gracefully")
sErr := e.Shutdown()
if sErr != nil {
log.Errorln(sErr)
log.Error("error shutting down", log.Ctx{"error": sErr})
}
}
......@@ -95,7 +99,7 @@ func main() {
candy.Must(err)
go func() {
log.Infof("start pserver at port %d", *port)
log.Info("serving pserver", log.Ctx{"port": *port})
err = http.Serve(l, nil)
candy.Must(err)
}()
......
go/glide.lock
浏览文件 @ 06c7686a
hash: 328e7b9b7306b45e7b9879139a9f86698115981f6283032e1312093a6a6ddb04
updated: 2017-10-16T08:00:23.484693528Z
hash: 107c058cf5c9163a75d40eef2273a793c36112683c25d72aa8288827fdde3a19
updated: 2017-10-30T03:46:19.137696069Z
imports:
- name: github.com/alecthomas/gometalinter
version: bae2f1293d092fd8167939d5108d1b025eaef9de
......@@ -99,6 +99,8 @@ imports:
version: d2709f9f1f31ebcda9651b03077758c1f3a0018c
- name: github.com/ghodss/yaml
version: 0ca9ea5df5451ffdf184b4428c902747c2c11cd7
- name: github.com/go-stack/stack
version: 817915b46b97fd7bb80e8ab6b69f01a53ac3eebf
- name: github.com/gogo/protobuf
version: 909568be09de550ed094403c2bf8a261b5bb730a
subpackages:
......@@ -120,8 +122,14 @@ imports:
- runtime
- runtime/internal
- utilities
- name: github.com/inconshreveable/log15
version: 0decfc6c20d9ca0ad143b0e89dcaa20f810b4fb3
- name: github.com/jonboulle/clockwork
version: 2eee05ed794112d45db504eb05aa693efd2b8b09
- name: github.com/mattn/go-colorable
version: 5411d3eea5978e6cdc258b30de592b60df6aba96
- name: github.com/mattn/go-isatty
version: 57fdcb988a5c543893cc61bce354a6e24ab70022
- name: github.com/matttproud/golang_protobuf_extensions
version: c12348ce28de40eed0136aa2b644d0ee0650e56c
subpackages:
......@@ -179,11 +187,12 @@ imports:
- lex/httplex
- trace
- name: golang.org/x/sys
version: 0f826bdd13b500be0f1d4004938ad978fcc6031e
version: e48874b42435b4347fc52bdee0424a52abc974d7
repo: https://github.com/golang/sys.git
vcs: git
subpackages:
- unix
- windows
- name: golang.org/x/text
version: 836efe42bb4aa16aaa17b9c155d8813d336ed720
repo: https://github.com/golang/text.git
......@@ -222,4 +231,3 @@ testImports:
version: 05e8a0eda380579888eb53c394909df027f06991
subpackages:
- assert
go/glide.yaml
浏览文件 @ 06c7686a
......@@ -26,3 +26,8 @@ import:
version: v1.1.0
- package: github.com/alecthomas/gometalinter
version: v1.2.1
- package: github.com/inconshreveable/log15
version: v2.13
- package: github.com/go-stack/stack
version: v1.6.0
- package: github.com/golang/protobuf
go/master/c/client.go
浏览文件 @ 06c7686a
......@@ -35,13 +35,19 @@ import (
"unsafe"
"github.com/PaddlePaddle/Paddle/go/master"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
)
var mu sync.Mutex
var handleMap = make(map[C.paddle_master_client]*master.Client)
var curHandle C.paddle_master_client
func init() {
log.Root().SetHandler(
log.LvlFilterHandler(log.LvlWarn, log.CallerStackHandler("%+v", log.StderrHandler)),
)
}
func add(c *master.Client) C.paddle_master_client {
mu.Lock()
defer mu.Unlock()
......@@ -117,7 +123,8 @@ func paddle_set_dataset(client C.paddle_master_client, path **C.char, size C.int
}
err := c.SetDataset(paths)
if err != nil {
log.Errorln(err)
log.Error("error set dataset",
log.Ctx{"error": err, "paths": paths})
return C.PADDLE_MASTER_ERROR
}
......@@ -167,7 +174,7 @@ func paddle_request_save_model(client C.paddle_master_client, trainerID string,
c := get(client)
need, err := c.RequestSaveModel(trainerID, time.Duration(blockMS)*time.Millisecond)
if err != nil {
log.Errorln(err)
log.Error("error request save model", log.Ctx{"error": err})
return C.PADDLE_MASTER_ERROR
}
......
go/master/client.go
浏览文件 @ 06c7686a
......@@ -21,7 +21,7 @@ import (
"github.com/PaddlePaddle/Paddle/go/connection"
"github.com/PaddlePaddle/recordio"
"github.com/coreos/etcd/clientv3"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
)
// Client is the client of the master server.
......@@ -75,7 +75,7 @@ func WithEtcd(endpoints []string, timeout time.Duration) func(*Client) error {
for {
err := f()
if err != nil {
log.Warningln(err)
log.Warn("create etcd client error", log.Ctx{"error": err})
} else {
break
}
......@@ -121,6 +121,7 @@ func (c *Client) StartGetRecords(passID int) {
}
func (c *Client) getRecords(passID int) {
i := 0
for {
t, err := c.getTask(passID)
if err != nil {
......@@ -130,18 +131,26 @@ func (c *Client) getRecords(passID int) {
c.ch <- record{nil, err}
break
}
if err.Error() == ErrPassAfter.Error() {
// wait util last pass finishes
time.Sleep(time.Second * 3)
continue
if i%60 == 0 {
log.Debug("getTask of passID error.",
log.Ctx{"error": err, "passID": passID})
i = 0
}
log.Errorf("getTask error: %s", err)
// if err.Error() == ErrPassAfter.Error()
// wait util last pass finishes
// if other error such as network error
// wait to reconnect or task time out
time.Sleep(time.Second * 3)
i += 3
continue
}
for _, chunk := range t.Chunks {
f, e := os.Open(chunk.Path)
if e != nil {
log.Errorln(e)
log.Error("error open chunk", log.Ctx{"error": e})
continue
}
......@@ -152,12 +161,15 @@ func (c *Client) getRecords(passID int) {
if s.Err() != nil {
c.ch <- record{nil, s.Err()}
log.Errorln(err, chunk.Path)
log.Error(
"error scan chunk",
log.Ctx{"error": err, "path": chunk.Path},
)
}
err = f.Close()
if err != nil {
log.Errorln(err)
log.Error("error close record file", log.Ctx{"error": err})
}
}
......@@ -166,7 +178,7 @@ func (c *Client) getRecords(passID int) {
// correct, but a reasonable approximation.
err = c.taskFinished(t.Meta.ID)
if err != nil {
log.Errorln(err)
log.Error("task finish callback error.", log.Ctx{"error": err})
}
}
}
......@@ -179,12 +191,12 @@ func (c *Client) monitorMaster(addrCh <-chan string) {
if curMaster == "" {
err := c.conn.Close()
if err != nil {
log.Errorln(err)
log.Error("close old master addr error", log.Ctx{"error": err})
}
} else {
err := c.conn.Connect(curMaster)
if err != nil {
log.Errorln(err)
log.Error("connect to new master addr error", log.Ctx{"error": err})
// connect to addr failed, set
// to last known addr in order
......
go/master/client_internal_test.go
浏览文件 @ 06c7686a
......@@ -25,8 +25,6 @@ import (
"testing"
"time"
log "github.com/sirupsen/logrus"
"github.com/PaddlePaddle/Paddle/go/connection"
"github.com/PaddlePaddle/recordio"
)
......@@ -36,10 +34,6 @@ const (
chunkPerTask = 10
)
func init() {
log.SetLevel(log.ErrorLevel)
}
func TestGetFinishTask(t *testing.T) {
const path = "/tmp/master_client_test_0"
......
go/master/client_test.go
浏览文件 @ 06c7686a
......@@ -117,6 +117,7 @@ func TestNextRecord(t *testing.T) {
if e != nil {
panic(e)
}
// test for n passes
for pass := 0; pass < 10; pass++ {
c.StartGetRecords(pass)
......
go/master/etcd_client.go
浏览文件 @ 06c7686a
......@@ -20,7 +20,7 @@ import (
"github.com/coreos/etcd/clientv3"
"github.com/coreos/etcd/clientv3/concurrency"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
)
const (
......@@ -44,7 +44,7 @@ type EtcdClient struct {
// NewEtcdClient creates a new EtcdClient.
func NewEtcdClient(endpoints []string, addr string, lockPath, addrPath, statePath string, ttlSec int) (*EtcdClient, error) {
log.Debugf("Connecting to etcd at %v", endpoints)
log.Debug("Connecting to etcd", log.Ctx{"endpoint": endpoints})
cli, err := clientv3.New(clientv3.Config{
Endpoints: endpoints,
DialTimeout: dialTimeout,
......@@ -64,12 +64,12 @@ func NewEtcdClient(endpoints []string, addr string, lockPath, addrPath, statePat
// one master running, but split-brain problem may cause
// multiple master servers running), and the cluster management
// software will kill one of them.
log.Infof("Trying to acquire lock at %s.", lockPath)
log.Info("Trying to acquire lock.", log.Ctx{"path": lockPath})
err = lock.Lock(context.TODO())
if err != nil {
return nil, err
}
log.Infof("Successfully acquired lock at %s.", lockPath)
log.Info("Successfully acquired lock at %s.", log.Ctx{"path": lockPath})
put := clientv3.OpPut(addrPath, addr)
resp, err := cli.Txn(context.Background()).If(lock.IsOwner()).Then(put).Commit()
......@@ -78,7 +78,8 @@ func NewEtcdClient(endpoints []string, addr string, lockPath, addrPath, statePat
}
if !resp.Succeeded {
log.Fatal("No longer owns the master lock. Exiting.")
log.Crit("No longer owns the master lock. Exiting.")
panic("No longer owns the master lock. Exiting.")
}
e := &EtcdClient{
......@@ -102,7 +103,7 @@ func (e *EtcdClient) Save(state []byte) error {
}
if !resp.Succeeded {
log.Errorln("No longer owns the lock, trying to lock again")
log.Error("No longer owns the lock, trying to lock again")
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
err := e.lock.Lock(ctx)
cancel()
......@@ -116,9 +117,10 @@ func (e *EtcdClient) Save(state []byte) error {
// to kill current master server. The current
// state is not saved, but the trainer's RPC
// call will fail, so the trainer will retry.
log.Fatalf("Could not acquire the lock at %s: %v. Exiting.", e.lockPath, err)
log.Crit("Could not acquire the lock at %s: %v. Exiting.", log.Ctx{"path": e.lockPath, "error": err})
panic("Could not acquire the lock at %s: %v. Exiting.")
}
log.Infof("Successfully acquired lock at %s.", e.lockPath)
log.Info("Successfully acquired lock at %s.", e.lockPath)
return e.Save(state)
}
......@@ -136,7 +138,7 @@ func (e *EtcdClient) Load() ([]byte, error) {
}
if !resp.Succeeded {
log.Errorln("No longer owns the lock, trying to lock and load again.")
log.Error("No longer owns the lock, trying to lock and load again.")
err = e.lock.Lock(context.Background())
if err != nil {
return nil, err
......@@ -163,7 +165,7 @@ func (e *EtcdClient) Shutdown() error {
if err == nil {
err = newErr
} else {
log.Errorln(newErr)
log.Error("shutdown error", log.Ctx{"error": newErr})
}
}
......@@ -192,7 +194,7 @@ func watchKey(c *clientv3.Client, key string, valChan chan<- string) {
for wresp := range rch {
for _, ev := range wresp.Events {
// if received event is DELETE, the value will be an empty string
log.Infof("received event %s, %q : %q\n", ev.Type, ev.Kv.Key, ev.Kv.Value)
log.Info("received event.", log.Ctx{"type": ev.Type, "key": ev.Kv.Key, "value": ev.Kv.Value})
valChan <- string(ev.Kv.Value)
}
}
......
go/master/service.go
浏览文件 @ 06c7686a
......@@ -25,7 +25,7 @@ import (
"sync"
"time"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
"github.com/PaddlePaddle/recordio"
)
......@@ -170,11 +170,11 @@ func (s *Service) recover() (bool, error) {
}
if state == nil {
log.Infoln("No state exists, not recovered.")
log.Info("No state exists, not recovered.")
return false, nil
}
log.Infof("Loaded snapshot of size: %d bytes.", len(state))
log.Info("Loaded snapshot.", log.Ctx{"size": len(state)})
gr, err := gzip.NewReader(bytes.NewReader(state))
if err != nil {
return false, err
......@@ -191,11 +191,11 @@ func (s *Service) recover() (bool, error) {
if err != nil {
// Only close failed, recover actually succeed, so
// just log error.
log.Errorln(err)
log.Error("error close recover file.", log.Ctx{"error": err})
}
s.state = tqs
log.WithFields(s.logFields()).Infof("Master recovered from snapshot, scheduling pending task timeout check.")
log.Info("Master recovered from snapshot, scheduling pending task timeout check.", s.logCtx())
for _, t := range s.state.Pending {
time.AfterFunc(s.timeoutDur, s.checkTimeoutFunc(t.Task.Meta.ID, t.Task.Meta.Epoch))
}
......@@ -224,7 +224,7 @@ func (s *Service) snapshot() error {
}
state := buf.Bytes()
log.Infof("Saving snapshot of size: %d bytes.", len(state))
log.Info("Saving snapshot.", log.Ctx{"size bytes": len(state)})
return s.store.Save(state)
}
......@@ -260,7 +260,7 @@ func readChunks(globPaths []string) ([]Chunk, error) {
}
count := index.NumChunks()
log.Infof("readChunks: file %s has %d chunks", path, count)
log.Info("reading chunks.", log.Ctx{"path": path, "num chunks": count})
for i := 0; i < count; i++ {
chunk := Chunk{
Path: path,
......@@ -300,7 +300,7 @@ func (s *Service) SetDataset(globPaths []string, _ *int) error {
err = s.snapshot()
if err != nil {
log.Errorln(err)
log.Error("snapshot error", log.Ctx{"error": err})
return err
}
close(s.ready)
......@@ -320,7 +320,7 @@ func (s *Service) processFailedTask(t taskEntry, epoch int) {
defer func() {
err := s.snapshot()
if err != nil {
log.Errorln(err)
log.Error("snapshot error", log.Ctx{"error": err})
}
}()
......@@ -328,12 +328,12 @@ func (s *Service) processFailedTask(t taskEntry, epoch int) {
t.NumFailure++
if t.NumFailure > s.failureMax {
log.Warningf("Task %v failed %d times, discard.", t.Task, t.NumFailure)
log.Warn("Task failed to many times, discard.", log.Ctx{"task": t.Task, "num failed": t.NumFailure})
s.state.Failed = append(s.state.Failed, t)
return
}
log.Warningf("Task %v failed %d times, re-dispatch.", t.Task, t.NumFailure)
log.Warn("Task failed, re-dispatch.", log.Ctx{"task": t.Task, "num failed": t.NumFailure})
s.state.Todo = append(s.state.Todo, t)
return
}
......@@ -353,8 +353,8 @@ func (s *Service) checkTimeoutFunc(taskID int, epoch int) func() {
}
// must be called with lock held.
func (s *Service) logFields() log.Fields {
return log.Fields{
func (s *Service) logCtx() log.Ctx {
return log.Ctx{
"todoLen": len(s.state.Todo),
"pendingLen": len(s.state.Pending),
"doneLen": len(s.state.Done),
......@@ -383,10 +383,10 @@ func (s *Service) GetTask(passID int, task *Task) error {
if len(s.state.Todo) == 0 {
if len(s.state.Done) == 0 && len(s.state.Pending) == 0 {
log.WithFields(s.logFields()).Warningln("All tasks failed, may start next pass")
log.Warn("All tasks failed, may start next pass", s.logCtx())
return ErrAllTaskFailed
}
log.WithFields(s.logFields()).Warningln("No more available task.")
log.Warn("No more available task.", s.logCtx())
return ErrNoMoreAvailable
}
......@@ -400,8 +400,9 @@ func (s *Service) GetTask(passID int, task *Task) error {
}
*task = t.Task
log.WithFields(s.logFields()).Infof("Task #%v dispatched.", t.Task.Meta)
ctx := s.logCtx()
ctx["task meta"] = t.Task.Meta
log.Info("Task dispatched.", ctx)
time.AfterFunc(s.timeoutDur, s.checkTimeoutFunc(t.Task.Meta.ID, t.Task.Meta.Epoch))
return nil
}
......@@ -417,7 +418,9 @@ func (s *Service) TaskFinished(taskID int, dummy *int) error {
t, ok := s.state.Pending[taskID]
if !ok {
log.WithFields(s.logFields()).Warningln("Pending task #%d not found.", taskID)
ctx := s.logCtx()
ctx["task id"] = taskID
log.Warn("Pending task not found.", ctx)
return nil
}
......@@ -426,7 +429,9 @@ func (s *Service) TaskFinished(taskID int, dummy *int) error {
s.state.Done = append(s.state.Done, t)
delete(s.state.Pending, taskID)
log.WithFields(s.logFields()).Infof("Task #%d finished.", taskID)
ctx := s.logCtx()
ctx["task id"] = taskID
log.Info("Task finished.", ctx)
if len(s.state.Todo) == 0 && len(s.state.Pending) == 0 {
// increase master side pass count if all tasks finished
s.state.CurPass++
......@@ -434,12 +439,14 @@ func (s *Service) TaskFinished(taskID int, dummy *int) error {
s.state.Done = []taskEntry{}
// TODO(typhoonzero): deal with failed tasks
s.state.Failed = []taskEntry{}
log.WithFields(s.logFields()).Warningf("all task finished, add new pass data, newpass: %d.", s.state.CurPass)
ctx := s.logCtx()
ctx["new pass"] = s.state.CurPass
log.Warn("all task finished, add new pass data.", ctx)
}
err := s.snapshot()
if err != nil {
log.Errorln(err)
log.Error("snapshot error", log.Ctx{"error": err})
}
return err
}
......@@ -455,7 +462,7 @@ func (s *Service) TaskFailed(meta TaskMeta, dummy *int) error {
t, ok := s.state.Pending[meta.ID]
if !ok {
log.WithFields(s.logFields()).Warningln("TaskFailed:Pending task #%v not found.", t.Task.Meta)
log.Warn("TaskFailed:Pending task not found.", log.Ctx{"task": t.Task.Meta})
return nil
}
......
go/proto/.gitignore 0 → 100644
浏览文件 @ 06c7686a
# Ignore everything in this directory
*
# Except this file
!.gitignore
go/pserver/CMakeLists.txt
浏览文件 @ 06c7686a
......@@ -13,5 +13,5 @@
# limitations under the License.
#
if(WITH_TESTING)
go_test(pserver_test DEPS paddle_go_optimizer)
go_test(pserver_test DEPS paddle_go_optimizer gen_proto_go)
endif()
go/pserver/client/c/cclient.go
浏览文件 @ 06c7686a
......@@ -45,9 +45,15 @@ import (
"github.com/PaddlePaddle/Paddle/go/pserver"
"github.com/PaddlePaddle/Paddle/go/pserver/client"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
)
func init() {
log.Root().SetHandler(
log.LvlFilterHandler(log.LvlWarn, log.CallerStackHandler("%+v", log.StderrHandler)),
)
}
var mu sync.Mutex
var handleMap = make(map[C.paddle_pserver_client]*client.Client)
var curHandle C.paddle_pserver_client
......@@ -164,10 +170,13 @@ func paddle_init_param(client C.paddle_pserver_client, param C.paddle_parameter,
if err != nil {
if err.Error() == pserver.AlreadyInitialized {
log.Warningf("parameter %s already initialized, treat paddle_init_param as successful.", name)
log.Warn(
"parameter already initialized, treat paddle_init_param as successful.",
log.Ctx{"parameter": name},
)
return C.PSERVER_OK
}
log.Errorln(err)
log.Error("error init param", log.Ctx{"error": err})
return C.PSERVER_ERROR
}
......@@ -180,11 +189,11 @@ func paddle_finish_init_params(client C.paddle_pserver_client) C.int {
err := c.FinishInitParams()
if err != nil {
if err.Error() == pserver.AlreadyInitialized {
log.Warningln("parameters already initialized, treat paddle_finish_init_params as successful.")
log.Warn("parameters already initialized, treat paddle_finish_init_params as successful.")
return C.PSERVER_OK
}
log.Errorln(err)
log.Error("error finish init params", log.Ctx{"error": err})
return C.PSERVER_ERROR
}
......@@ -205,7 +214,7 @@ func paddle_send_grads(client C.paddle_pserver_client, grads **C.paddle_gradient
c := get(client)
err := c.SendGrads(gs)
if err != nil {
log.Errorln(err)
log.Error("error send grads", log.Ctx{"error": err})
return C.PSERVER_ERROR
}
......@@ -222,7 +231,7 @@ func paddle_get_params(client C.paddle_pserver_client, dst **C.paddle_parameter,
c := get(client)
ps, err := c.GetParams(ns)
if err != nil {
log.Errorln(err)
log.Error("error get params", log.Ctx{"error": err})
return C.PSERVER_ERROR
}
......@@ -231,7 +240,13 @@ func paddle_get_params(client C.paddle_pserver_client, dst **C.paddle_parameter,
for i, p := range ps {
pn[i] = p.Name
}
log.Errorf("pserver returned wrong number of parameters. Requested: %s, returned: %s.", strings.Join(pn, ", "), strings.Join(ns, ", "))
log.Error(
"pserver returned wrong number of parameters.",
log.Ctx{
"Requested": strings.Join(pn, ", "),
"Returned": strings.Join(ns, ", "),
},
)
return C.PSERVER_ERROR
}
......@@ -241,7 +256,13 @@ func paddle_get_params(client C.paddle_pserver_client, dst **C.paddle_parameter,
for i, p := range ps {
pn[i] = p.Name
}
log.Errorf("pserver returned wrong parameters, or not in requested order. Requested: %s, returned: %s.", strings.Join(pn, ", "), strings.Join(ns, ", "))
log.Error(
"pserver returned wrong parameters, or not in requested order.",
log.Ctx{
"Requested": strings.Join(pn, ", "),
"Returned": strings.Join(ns, ", "),
},
)
return C.PSERVER_ERROR
}
}
......@@ -251,13 +272,19 @@ func paddle_get_params(client C.paddle_pserver_client, dst **C.paddle_parameter,
param := *(**C.paddle_parameter)(unsafe.Pointer((uintptr(unsafe.Pointer(dst)) + uintptr(i)*unsafe.Sizeof(*dst))))
if unsafe.Pointer(param) == nil {
log.Errorln("must pre-allocate parameter.")
log.Error("must pre-allocate parameter.")
return C.PSERVER_ERROR
}
if unsafe.Pointer(param.content) != nil {
if int(param.content_len) != len(p.Content) {
log.Errorf("the pre-allocated content len does not match parameter content len. Pre-allocated len: %d, returned len: %d", param.content_len, len(p.Content))
log.Error(
"the pre-allocated content len does not match parameter content len.",
log.Ctx{
"Pre-allocated len": param.content_len,
"Returned len": len(p.Content),
},
)
return C.PSERVER_ERROR
}
}
......
go/pserver/client/client.go
浏览文件 @ 06c7686a
......@@ -22,7 +22,7 @@ import (
"github.com/PaddlePaddle/Paddle/go/connection"
"github.com/PaddlePaddle/Paddle/go/pserver"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
)
// TODO(helin): add RPC call retry logic
......@@ -84,7 +84,7 @@ func (c *Client) monitorPservers(l Lister, pserverNum int) {
if curServers[i].Addr == "" {
err := c.pservers[i].Close()
if err != nil {
log.Errorln(err)
log.Error("error closing connection to pserver", log.Ctx{"error": err})
}
continue
......@@ -92,7 +92,7 @@ func (c *Client) monitorPservers(l Lister, pserverNum int) {
err := c.pservers[i].Connect(curServers[i].Addr)
if err != nil {
log.Errorln(err)
log.Error("error connecting to pserver", log.Ctx{"error": err})
// connect to addr failed, set
// to last known addr in order
......@@ -137,7 +137,7 @@ func (c *Client) FinishInitParams() error {
return err
}
}
return nil
return c.sel.Done()
}
// SendGrads sends gradients to parameter servers for updating
......
go/pserver/client/client_test.go
浏览文件 @ 06c7686a
......@@ -30,7 +30,7 @@ import (
"github.com/PaddlePaddle/Paddle/go/pserver"
"github.com/PaddlePaddle/Paddle/go/pserver/client"
"github.com/coreos/etcd/clientv3"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
)
const (
......@@ -90,7 +90,7 @@ func initEtcdClient() {
DialTimeout: time.Second * time.Duration(1),
})
if err != nil {
log.Errorf("err %v", err)
log.Error("error init etcd client", log.Ctx{"error": err})
}
ctx, cancel := context.WithTimeout(context.Background(), timeout)
_, err = client.Delete(ctx, pserver.PsDesired)
......
go/pserver/client/etcd_client.go
浏览文件 @ 06c7686a
此差异已折叠。
go/pserver/etcd_client.go
浏览文件 @ 06c7686a
此差异已折叠。
go/pserver/optimizer.go
浏览文件 @ 06c7686a
......@@ -25,7 +25,7 @@ import (
"fmt"
"unsafe"
log "github.com/sirupsen/logrus"
log "github.com/inconshreveable/log15"
)
type optimizer struct {
......@@ -56,12 +56,12 @@ func newOptimizer(paramWithConfigs ParameterWithConfig, State []byte) *optimizer
c := paramWithConfigs.Config
s := State
paramBufferSize := C.size_t(len(p.Content))
log.WithFields(log.Fields{
log.Info("New Optimizer Created with config", log.Ctx{
"ElementType": p.ElementType,
"ParamSize": paramBufferSize,
"ConfigSize": len(c),
"StateSize": len(s),
}).Info("New Optimizer Created with config:")
})
var cbuffer unsafe.Pointer
cbuffer = C.malloc(paramBufferSize)
......@@ -71,22 +71,41 @@ func newOptimizer(paramWithConfigs ParameterWithConfig, State []byte) *optimizer
cstate = unsafe.Pointer(&s[0])
}
var cptr (*C.uchar)
if len(c) > 0 {
cptr = (*C.uchar)(&c[0])
} else {
log.Error("empty config", "param name", paramWithConfigs.Param.Name)
}
o.config = c
o.opt = C.paddle_create_optimizer((*C.uchar)(&c[0]), C.int(len(c)),
C.paddle_element_type(p.ElementType), cbuffer, C.int(paramBufferSize), (*C.char)(cstate), C.int(len(s)))
o.opt = C.paddle_create_optimizer(
cptr,
C.int(len(c)),
C.paddle_element_type(p.ElementType),
cbuffer,
C.int(paramBufferSize),
(*C.char)(cstate),
C.int(len(s)),
)
return o
}
func (o *optimizer) GetWeights() []byte {
var buffer unsafe.Pointer
// we do not own the buffer, no need to free later.
bufferLen := C.paddle_optimizer_get_weights(o.opt, &buffer)
return cArrayToSlice(buffer, int(bufferLen)*C.sizeof_float)
}
func (o *optimizer) GetStates() []byte {
var cbuffer *C.char
// we owns the state buffer, need to free later.
cbufferLen := C.paddle_optimizer_get_state(o.opt, &cbuffer)
return cArrayToSlice(unsafe.Pointer(cbuffer), int(cbufferLen))
buf := cArrayToSlice(unsafe.Pointer(cbuffer), int(cbufferLen))
cpy := make([]byte, len(buf))
copy(cpy, buf)
C.free(unsafe.Pointer(cbuffer))
return cpy
}
func (o *optimizer) UpdateParameter(g Gradient) error {
......
go/pserver/optimizer_test.go
浏览文件 @ 06c7686a
此差异已折叠。
go/pserver/service.go
浏览文件 @ 06c7686a
此差异已折叠。
go/pserver/service_internal_test.go 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
go/pserver/service_test.go
浏览文件 @ 06c7686a
此差异已折叠。
paddle/capi/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/capi/Matrix.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/capi/gradient_machine.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/capi/matrix.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/capi/tests/test_Matrix.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/cuda/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/cuda/include/hl_cnn.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/cuda/include/hl_matrix.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/cuda/src/hl_cuda_cnn.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/cuda/src/hl_cuda_matrix.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/attribute.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/attribute.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/backward.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/backward_test.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/block_desc.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/block_desc.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/data_type.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/ddim.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/ddim.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/executor.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/executor.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/framework.proto
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/lod_rank_table.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/lod_rank_table.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/lod_tensor.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/lod_tensor.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/lod_tensor.md
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/lod_tensor_array.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/lod_tensor_test.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/lod_tensor_test.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/op_desc.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/op_desc.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/op_info.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/op_proto_maker.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/op_registry.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/op_registry.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/operator.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/operator.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/operator_test.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/program_desc.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/program_desc.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/program_desc_test.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/proto_desc.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/prune.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/prune_test.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/scope.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/scope.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/scope_test.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/selected_rows.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/shape_inference.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/shape_inference.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/tensor.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/tensor_array.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/tensor_array.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/tensor_impl.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/tensor_test.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/type_defs.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/var_desc.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/var_desc.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/var_type.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/framework/variable.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/ConvOp.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/ConvOpTest.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/FunctionTest.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/GemmConvOp.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/Im2Col.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/Im2ColOp.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/Im2ColOpGpu.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/Im2ColTest.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/ScaleSubRegionOp.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/ScaleSubRegionOp.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/ScaleSubRegionOpGpu.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/function/ScaleSubRegionOpTest.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/CRFLayer.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/Layer.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MKLDNNAddtoLayer.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MKLDNNAddtoLayer.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MKLDNNBase.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MKLDNNBatchNormLayer.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MKLDNNBatchNormLayer.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MKLDNNLayer.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MKLDNNLayer.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MaxPoolWithMaskLayer.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/MaxPoolWithMaskLayer.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/PoolLayer.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/ROIPoolLayer.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/ROIPoolLayer.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/ScaleSubRegionLayer.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/layers/ScaleSubRegionLayer.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/tests/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/tests/MKLDNNTester.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/tests/test_Expand.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/gserver/tests/test_MaxPoolingWithMaskOutput.cpp 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/BaseMatrix.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/CpuSparseMatrix.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/MKLDNNMatrix.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/MKLDNNMatrix.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/MathFunctions.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/MathFunctions.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/Matrix.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/Matrix.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/RowBuffer.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/SparseMatrix.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/SparseRowMatrix.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/Vector.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/Vector.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/tests/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/math/tests/TensorCheck.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/memory/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/memory/detail/meta_cache.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/memory/memcpy.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/memory/memory.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/accuracy_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/accuracy_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/accuracy_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/activation_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/activation_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/activation_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/adadelta_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/adagrad_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/adam_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/adam_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/adamax_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/adamax_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/array_operator.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/array_to_lod_tensor_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/assign_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/auc_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/auc_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/batch_norm_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/batch_norm_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/batch_norm_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/batch_norm_op.md 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/beam_search_decode_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/beam_search_decode_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/beam_search_decode_op_test.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/bilinear_tensor_product_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/bilinear_tensor_product_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/bilinear_tensor_product_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/cast_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/cast_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/cast_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/chunk_eval_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/chunk_eval_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/clip_by_norm_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/clip_by_norm_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/clip_by_norm_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/clip_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/compare_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/compare_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/compare_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/concat_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/cond_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conditional_block_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv2d_transpose_cudnn_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv2d_transpose_cudnn_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv_cudnn_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv_cudnn_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv_shift_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv_shift_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv_transpose_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv_transpose_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/conv_transpose_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/cos_sim_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/crf_decoding_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/crf_decoding_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/crop_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/cross_entropy_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/dropout_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/dropout_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/elementwise_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/expand_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/expand_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/expand_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/fc_op.cc 已删除 100644 → 0
浏览文件 @ 7461b359
此差异已折叠。
paddle/operators/feed_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/fetch_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/fill_constant_batch_size_like_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/fill_constant_batch_size_like_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/gather_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/gru_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/gru_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/gru_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/gru_unit_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/gru_unit_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/huber_loss_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/huber_loss_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/huber_loss_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/images/batch_norm_fork.dot 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/images/batch_norm_fork.png 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/images/batch_norm_op_kernel.png 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/increment_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/interp_op.cc 已删除 100644 → 0
浏览文件 @ 7461b359
此差异已折叠。
paddle/operators/l1_norm_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/l1_norm_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/l1_norm_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/linear_chain_crf_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/linear_chain_crf_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/linear_chain_crf_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/load_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lod_array_length_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lod_rank_table_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lod_reset_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lod_reset_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lod_reset_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lod_tensor_to_array_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lookup_table_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lookup_table_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lookup_table_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lrn_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lrn_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lrn_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lstm_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lstm_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lstm_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lstm_unit_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lstm_unit_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/lstm_unit_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/context_project.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/context_project.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/context_project.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/CMakeLists.txt 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/activation_functions.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/avx_functions.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/gru_cpu_kernel.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/gru_gpu_kernel.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/gru_kernel.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/lstm_cpu_kernel.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/lstm_gpu_kernel.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/detail/lstm_kernel.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/gru_compute.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/gru_compute.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/gru_compute.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/im2col.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/im2col.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/im2col.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/lstm_compute.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/lstm_compute.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/lstm_compute.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/matmul.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/pooling.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/pooling.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/pooling.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/sequence2batch.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/sequence2batch.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/sequence2batch.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/sequence_pooling.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/sequence_pooling.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/math/sequence_pooling.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/matmul_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/matmul_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/matmul_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/mean_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/mean_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/merge_lod_tensor_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/minus_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/momentum_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/momentum_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/mul_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/mul_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/mul_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/multiplex_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/multiplex_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/name_convention.md
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/nccl/CMakeLists.txt 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/nccl/nccl_gpu_common.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/nccl/nccl_gpu_common.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/nccl_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/nccl_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/nccl_op_test.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/pad_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/pool_cudnn_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/pool_cudnn_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/pool_cudnn_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/pool_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/pool_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/positive_negative_pair_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/positive_negative_pair_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/precision_recall_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/precision_recall_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/prelu_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/proximal_adagrad_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/proximal_adagrad_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/proximal_adagrad_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/proximal_gd_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/proximal_gd_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/proximal_gd_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/rank_loss_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/recurrent_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/recurrent_op.h 已删除 100644 → 0
浏览文件 @ 7461b359
此差异已折叠。
paddle/operators/reduce_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/reduce_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/reshape_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/reshape_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/rmsprop_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/rnn_memory_helper_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/save_load_op_test.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/save_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/scale_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/scale_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/scale_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/scatter_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/scatter_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/scatter_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/seq_expand_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/seq_expand_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/seq_expand_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sequence_conv_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sequence_conv_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sequence_pool_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sgd_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sgd_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sgd_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/shrink_rnn_memory_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sign_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sign_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/softmax_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/softmax_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/softmax_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/split_lod_tensor_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/split_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/squared_l2_norm_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/squared_l2_norm_op.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/squared_l2_norm_op.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sum_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sum_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/sum_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/tensor_array_read_write_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/top_k_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/top_k_op.cu
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/top_k_op.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/transpose_op.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/operators/while_op.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/adam_optimizer.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/adam_optimizer.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/lr_policy.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/optimizer.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/optimizer.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/sgd_optimizer.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/sgd_optimizer.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/optimizer/tensor.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/parameter/Parameter.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/call_once.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/cudnn_helper.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/device_context.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/device_context.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/dynload/cublas.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/dynload/cudnn.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/dynload/nccl.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/dynload/nccl.h 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/enforce.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/environment_test.cc 已删除 100644 → 0
浏览文件 @ 7461b359
此差异已折叠。
paddle/platform/gpu_info.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/nccl_test.cu 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/place.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/platform/transform.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/pserver/ParameterClient2.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/pybind/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/pybind/print_operators_doc.cc 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/pybind/protobuf.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/pybind/pybind.cc
浏览文件 @ 06c7686a
此差异已折叠。
paddle/pybind/tensor_py.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/scripts/cluster_train_v2/fabric/conf.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/scripts/cluster_train_v2/fabric/run.sh 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
paddle/scripts/docker/build.sh
浏览文件 @ 06c7686a
此差异已折叠。
paddle/scripts/travis/build_doc.sh
浏览文件 @ 06c7686a
此差异已折叠。
paddle/testing/TestUtil.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/trainer/MergeModel.cpp
浏览文件 @ 06c7686a
此差异已折叠。
paddle/trainer/tests/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
paddle/utils/Excepts.h
浏览文件 @ 06c7686a
此差异已折叠。
paddle/utils/arch/osx/Excepts.cpp
浏览文件 @ 06c7686a
此差异已折叠。
proto/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
proto/ModelConfig.proto
浏览文件 @ 06c7686a
此差异已折叠。
proto/TrainerConfig.proto
浏览文件 @ 06c7686a
此差异已折叠。
python/CMakeLists.txt
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/utils/merge_model.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/__init__.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/dataset/common.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/dataset/imdb.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/__init__.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/backward.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/evaluator.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/executor.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/initializer.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/io.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/layers.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/net_drawer.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/nets.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/optimizer.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/regularizer.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/.gitignore 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/CMakeLists.txt 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/book/test_fit_a_line.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/book/test_word2vec.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_assign_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_auc_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_batch_norm_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_cast_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_chunk_eval_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_clip_by_norm_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_compare_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_conditional_block.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_conv3d_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_crf_decoding_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_executor_and_mul.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_expand_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_gru_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_huber_loss_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_inference_model_io.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_initializer.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_layers.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_lod_rank_table.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_lod_reset_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_lod_tensor_array.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_lrn_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_lstm_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_matmul_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_momentum_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_nccl_init_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_op_support_gpu.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_optimizer.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_recurrent_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_regularizer.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_seq_conv.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_seq_expand.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_sgd_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_shrink_rnn_memory.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_sign_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_squared_l2_norm_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/fluid/tests/test_while_op.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/framework/__init__.py 已删除 100644 → 0
浏览文件 @ 7461b359
此差异已折叠。
python/paddle/v2/framework/layers.py 已删除 100644 → 0
浏览文件 @ 7461b359
此差异已折叠。
python/paddle/v2/framework/math_ops.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/image.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/model.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/optimizer.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/parameters.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/plot/plot.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/reader/creator.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/tests/test_paramconf_order.py 0 → 100644
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/topology.py
浏览文件 @ 06c7686a
此差异已折叠。
python/paddle/v2/trainer.py
浏览文件 @ 06c7686a
此差异已折叠。
python/requirements.txt
浏览文件 @ 06c7686a
此差异已折叠。
python/setup.py.in
浏览文件 @ 06c7686a
此差异已折叠。
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或