Skip to content

P
Paddle

机器未来 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 a38a8db9 编写于 作者: F fengjiayi
浏览文件
操作

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

上级 10ec329b 0a641ba3
展开全部 隐藏空白更改
内联 并排
Showing with 51873 addition and 27333 deletion

要显示的变更太多。

To preserve performance only 1000 of 1000+ files are displayed.
文本差异 电子邮件补丁
.clang-format
浏览文件 @ a38a8db9
......@@ -19,7 +19,7 @@ BasedOnStyle: Google
IndentWidth: 2
TabWidth: 2
ContinuationIndentWidth: 4
AccessModifierOffset: -2 # The private/protected/public has no indent in class
AccessModifierOffset: -1 # The private/protected/public has no indent in class
Standard: Cpp11
AllowAllParametersOfDeclarationOnNextLine: true
BinPackParameters: false
......
.copyright.hook 已删除 100644 → 0
浏览文件 @ 10ec329b
from __future__ import absolute_import
from __future__ import print_function
from __future__ import unicode_literals
import argparse
import io, re
import sys, os
import subprocess
import platform
COPYRIGHT = '''
Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
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.
'''
LANG_COMMENT_MARK = None
NEW_LINE_MARK = None
COPYRIGHT_HEADER = None
if platform.system() == "Windows":
NEW_LINE_MARK = "\r\n"
else:
NEW_LINE_MARK = '\n'
COPYRIGHT_HEADER = COPYRIGHT.split(NEW_LINE_MARK)[1]
p = re.search('(\d{4})', COPYRIGHT_HEADER).group(0)
process = subprocess.Popen(["date", "+%Y"], stdout=subprocess.PIPE)
date, err = process.communicate()
date = date.decode("utf-8").rstrip("\n")
COPYRIGHT_HEADER = COPYRIGHT_HEADER.replace(p, date)
def generate_copyright(template, lang='C'):
if lang == 'Python':
LANG_COMMENT_MARK = '#'
else:
LANG_COMMENT_MARK = "//"
lines = template.split(NEW_LINE_MARK)
BLANK = " "
ans = LANG_COMMENT_MARK + BLANK + COPYRIGHT_HEADER + NEW_LINE_MARK
for lino, line in enumerate(lines):
if lino == 0 or lino == 1 or lino == len(lines) - 1: continue
if len(line) == 0:
BLANK = ""
else:
BLANK = " "
ans += LANG_COMMENT_MARK + BLANK + line + NEW_LINE_MARK
return ans + "\n"
def lang_type(filename):
if filename.endswith(".py"):
return "Python"
elif filename.endswith(".h"):
return "C"
elif filename.endswith(".c"):
return "C"
elif filename.endswith(".hpp"):
return "C"
elif filename.endswith(".cc"):
return "C"
elif filename.endswith(".cpp"):
return "C"
elif filename.endswith(".cu"):
return "C"
elif filename.endswith(".cuh"):
return "C"
elif filename.endswith(".go"):
return "C"
elif filename.endswith(".proto"):
return "C"
else:
print("Unsupported filetype %s", filename)
exit(0)
PYTHON_ENCODE = re.compile("^[ \t\v]*#.*?coding[:=][ \t]*([-_.a-zA-Z0-9]+)")
def main(argv=None):
parser = argparse.ArgumentParser(
description='Checker for copyright declaration.')
parser.add_argument('filenames', nargs='*', help='Filenames to check')
args = parser.parse_args(argv)
retv = 0
for filename in args.filenames:
fd = io.open(filename, encoding="utf-8")
first_line = fd.readline()
second_line = fd.readline()
if "COPYRIGHT (C)" in first_line.upper(): continue
if first_line.startswith("#!") or PYTHON_ENCODE.match(
second_line) != None or PYTHON_ENCODE.match(first_line) != None:
continue
original_contents = io.open(filename, encoding="utf-8").read()
new_contents = generate_copyright(
COPYRIGHT, lang_type(filename)) + original_contents
print('Auto Insert Copyright Header {}'.format(filename))
retv = 1
with io.open(filename, 'w') as output_file:
output_file.write(new_contents)
return retv
if __name__ == '__main__':
exit(main())
.pre-commit-config.yaml
浏览文件 @ a38a8db9
......@@ -23,7 +23,7 @@ repos:
- id: clang-format-with-version-check
name: clang-format
description: Format files with ClangFormat.
entry: bash ./.clang_format.hook -i
entry: bash ./tools/codestyle/clang_format.hook -i
language: system
files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|proto)$
- repo: local
......@@ -34,6 +34,14 @@ repos:
entry: bash ./tools/codestyle/cpplint_pre_commit.hook
language: system
files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx)$
- repo: local
hooks:
- id: pylint-doc-string
name: pylint
description: Check python docstring style using docstring_checker.
entry: bash ./tools/codestyle/pylint_pre_commit.hook
language: system
files: \.(py)$
- repo: https://github.com/PaddlePaddle/pre-commit-golang
sha: 8337620115c25ff8333f1b1a493bd031049bd7c0
hooks:
......@@ -44,7 +52,7 @@ repos:
hooks:
- id: copyright_checker
name: copyright_checker
entry: python ./.copyright.hook
entry: python ./tools/codestyle/copyright.hook
language: system
files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|proto|py)$
exclude: (?!.*third_party)^.*$ | (?!.*book)^.*$
.travis.yml
浏览文件 @ a38a8db9
......@@ -18,6 +18,8 @@ env:
addons:
ssh_known_hosts: 13.229.163.131
before_install:
# For pylint dockstring checker
- sudo pip install pylint pytest astroid isort
- |
function timeout() { perl -e 'alarm shift; exec @ARGV' "$@"; }
script:
......@@ -25,15 +27,6 @@ script:
# 43min timeout
paddle/scripts/paddle_docker_build.sh ${JOB}
if [ $? -eq 0 ] || [ $? -eq 142 ]; then true; else exit 1; fi;
- |
if [[ "$JOB" != "doc" ]]; then exit 0; fi;
# For document only
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
......
AUTHORS.md
浏览文件 @ a38a8db9
......@@ -4,6 +4,7 @@
| backyes | Yan-Fei Wang |
| baiyfbupt | Yi-Fan Bai |
| beckett1124 | Bin Qi |
| ChengduoZH | Cheng-Duo Zhao|
| chengxiaohua1105 | Xiao-Hua Cheng |
| cxwangyi, yiwangbaidu, wangkuiyi | Yi Wang |
| cxysteven | Xing-Yi Cheng |
......@@ -21,6 +22,7 @@
| jczaja | Jacek Czaja |
| JiayiFeng | Jia-Yi Feng |
| kbinias | Krzysztof Binias |
| kexinzhao | Ke-Xin Zhao |
| kuke | Yi-Bing Liu |
| lcy-seso | Ying Cao |
| lipeng-unisound | Peng Li |
......@@ -44,6 +46,7 @@
| tianbingsz | Tian-Bing Xu |
| tpatejko | Tomasz Patejko |
| typhoonzero | Yi Wu |
| velconia | Qi-Yang Min |
| wanghaoshuang | Hao-Shuang Wang |
| wangyang59 | Yang Wang |
| wangzhen-nlp | Zhen Wang |
......
CMakeLists.txt
浏览文件 @ a38a8db9
......@@ -25,7 +25,6 @@ message(STATUS "CXX compiler: ${CMAKE_CXX_COMPILER}, version: "
message(STATUS "C compiler: ${CMAKE_C_COMPILER}, version: "
"${CMAKE_C_COMPILER_ID} ${CMAKE_C_COMPILER_VERSION}")
find_package(Sphinx)
if(NOT CMAKE_CROSSCOMPILING)
find_package(CUDA QUIET)
endif(NOT CMAKE_CROSSCOMPILING)
......@@ -42,7 +41,6 @@ option(WITH_MKL "Compile PaddlePaddle with MKL support." ${AVX_FO
option(WITH_DSO "Compile PaddlePaddle with dynamic linked CUDA" ON)
option(WITH_TESTING "Compile PaddlePaddle with unit testing" OFF)
option(WITH_SWIG_PY "Compile PaddlePaddle with inference api" ON)
option(WITH_STYLE_CHECK "Compile PaddlePaddle with style check" ON)
option(WITH_PYTHON "Compile PaddlePaddle with python interpreter" ON)
option(WITH_DOUBLE "Compile PaddlePaddle with double precision" OFF)
option(WITH_RDMA "Compile PaddlePaddle with RDMA support" OFF)
......@@ -57,10 +55,25 @@ option(WITH_FLUID_ONLY "Compile PaddlePaddle fluid only" OFF)
option(WITH_GOLANG "Compile PaddlePaddle with GOLANG" OFF)
option(GLIDE_INSTALL "Download and install go dependencies " ON)
option(USE_NNPACK "Compile PaddlePaddle with NNPACK library" OFF)
option(WITH_DISTRIBUTE "Compile with grpc distributed support" OFF)
option(WITH_DISTRIBUTE "Compile with distributed support" OFF)
option(USE_EIGEN_FOR_BLAS "Use matrix multiplication in Eigen" OFF)
option(EIGEN_USE_THREADS "Compile with multi-threaded Eigen" OFF)
option(WITH_ARM_FP16 "Use half precision support on armv8.2-a cpu" OFF)
option(WITH_FAST_BUNDLE_TEST "Bundle tests that can be run in a single process together to reduce launch overhead" OFF)
option(WITH_CONTRIB "Compile the third-party contributation" OFF)
option(REPLACE_ENFORCE_GLOG "Replace PADDLE_ENFORCE with glog/CHECK for better debug." OFF)
option(WITH_ANAKIN "Compile with Anakin library" OFF)
option(WITH_GRPC "Use grpc as the default rpc framework" ${WITH_DISTRIBUTE})
option(WITH_BRPC_RDMA "Use brpc rdma as the rpc protocal" OFF)
option(WITH_INFERENCE "Compile fluid inference library" ON)
option(WITH_SYSTEM_BLAS "Use system blas library" OFF)
option(PY_VERSION "Compile PaddlePaddle with python3 support" ${PY_VERSION})
# PY_VERSION
if(NOT PY_VERSION)
set(PY_VERSION 2.7)
endif()
set(PYBIND11_PYTHON_VERSION ${PY_VERSION})
# CMAKE_BUILD_TYPE
if(NOT CMAKE_BUILD_TYPE)
......@@ -98,9 +111,17 @@ if(ANDROID OR IOS)
add_definitions(-DPADDLE_MOBILE_INFERENCE)
endif()
if (APPLE OR WIN32)
set(WITH_MKL OFF CACHE STRING
"Disable MKL for building on mac and windows" FORCE)
endif()
set(THIRD_PARTY_PATH "${CMAKE_BINARY_DIR}/third_party" CACHE STRING
"A path setting third party libraries download & build directories.")
set(FLUID_INSTALL_DIR "${CMAKE_BINARY_DIR}/fluid_install_dir" CACHE STRING
"A path setting fluid shared and static libraries")
if (WITH_C_API AND WITH_PYTHON)
message(WARNING "It is suggest not embedded a python interpreter in Paddle "
"when using C-API. It will give an unpredictable behavior when using a "
......@@ -117,17 +138,30 @@ else()
set(THIRD_PARTY_BUILD_TYPE Release)
endif()
if(WITH_MKL)
option(MKL_SPLIT_GEMM "PaddlePaddle MKL gemm would split to small ones" OFF)
if (MKL_SPLIT_GEMM)
add_definitions(-DPADDLE_MKL_SPLIT_GEMM)
endif()
endif()
set(WITH_MKLML ${WITH_MKL})
if (WITH_MKL AND AVX2_FOUND)
set(WITH_MKLDNN ON)
else()
message(STATUS "Do not have AVX2 intrinsics and disabled MKL-DNN")
set(WITH_MKLDNN OFF)
if (NOT DEFINED WITH_MKLDNN)
if (WITH_MKL AND AVX2_FOUND)
set(WITH_MKLDNN ON)
else()
message(STATUS "Do not have AVX2 intrinsics and disabled MKL-DNN")
set(WITH_MKLDNN OFF)
endif()
endif()
if (REPLACE_ENFORCE_GLOG)
add_definitions("-DREPLACE_ENFORCE_GLOG")
endif()
########################################################################################
include(external/mklml) # download mklml package
include(external/xbyak) # download xbyak package
include(external/libxsmm) # download, build, install libxsmm
include(external/zlib) # download, build, install zlib
include(external/gflags) # download, build, install gflags
include(external/glog) # download, build, install glog
......@@ -143,17 +177,45 @@ include(external/any) # download libn::any
include(external/eigen) # download eigen3
include(external/pybind11) # download pybind11
include(external/cares)
include(external/grpc)
include(external/cub)
if(WITH_DISTRIBUTE)
if(WITH_GRPC)
include(external/grpc)
message(STATUS "Use grpc framework.")
else()
message(STATUS "Use brpc framework.")
include(external/leveldb)
include(external/brpc)
endif()
endif()
if(WITH_BRPC_RDMA)
message(STATUS "Use brpc with rdma.")
if(WITH_GRPC)
message(FATAL_ERROR "Can't use grpc with brpc rdma.")
endif()
if(NOT WITH_DISTRIBUTE)
message(FATAL_ERROR "Can't use brpc rdma in no distribute env.")
endif()
endif()
include(external/snappy) # download snappy
include(external/snappystream)
include(external/threadpool)
set(WITH_ANAKIN OFF CACHE STRING "Disable Anakin first, will add it later." FORCE)
if(WITH_GPU)
include(cuda)
include(tensorrt)
include(external/anakin)
endif()
include(cudnn) # set cudnn libraries, must before configure
include(cupti)
include(configure) # add paddle env configuration
include(generic) # simplify cmake module
include(package) # set paddle packages
include(cpplint) # set paddle c++ style
include(ccache) # set ccache for compilation
include(util) # set unittest and link libs
include(rdma) # set rdma libraries
......@@ -164,7 +226,7 @@ include(inference_lib) # add paddle fluid inference libraries
include_directories("${PADDLE_SOURCE_DIR}")
include_directories("${PADDLE_SOURCE_DIR}/paddle/cuda/include")
include_directories("${PADDLE_SOURCE_DIR}/paddle/legacy/cuda/include")
include_directories("${CMAKE_CURRENT_BINARY_DIR}/proto")
include_directories("${CMAKE_CURRENT_BINARY_DIR}/go/pserver/client/c")
......@@ -177,11 +239,6 @@ set(EXTERNAL_LIBS
${PYTHON_LIBRARIES}
)
if(WITH_GPU)
include(cuda)
include(tensorrt)
endif(WITH_GPU)
if(WITH_AMD_GPU)
find_package(HIP)
include(hip)
......@@ -191,6 +248,10 @@ if(WITH_MKLML)
list(APPEND EXTERNAL_LIBS ${MKLML_IOMP_LIB})
endif()
if(WITH_LIBXSMM)
list(APPEND EXTERNAL_LIBS ${LIBXSMM_LIBS})
endif()
if(WITH_MKLDNN)
list(APPEND EXTERNAL_LIBS ${MKLDNN_LIB})
endif()
......@@ -202,10 +263,10 @@ endif(USE_NNPACK)
add_subdirectory(proto)
if(NOT MOBILE_INFERENCE)
if(NOT MOBILE_INFERENCE AND NOT WITH_FLUID_ONLY)
# "add_subdirectory(go)" should be placed after the following loine,
# because it depends on paddle/optimizer.
add_subdirectory(paddle/optimizer)
add_subdirectory(paddle/legacy/optimizer)
endif()
# "add_subdirectory(paddle)" and "add_subdirectory(python)" should be
......@@ -226,5 +287,7 @@ if(WITH_PYTHON)
endif()
if(WITH_DOC)
find_package(Sphinx REQUIRED)
find_python_module(recommonmark REQUIRED)
add_subdirectory(doc)
endif()
CONTRIBUTING.md
浏览文件 @ a38a8db9
......@@ -58,6 +58,8 @@ PaddlePaddle uses this [Git branching model](http://nvie.com/posts/a-successful-
create mode 100644 233
```
NOTE: The `yapf` installed by `pip install pre-commit` and `conda install -c conda-forge pre-commit` is slightly different. Paddle developers use `pip install pre-commit`.
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).
......@@ -157,4 +159,4 @@ This will enable VLOG messages generated by `buddy_allocator.{h,cc}` and in the
- 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)
- verbose level 7: [math](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/legacy/math)
Dockerfile
浏览文件 @ a38a8db9
......@@ -23,13 +23,13 @@ ENV HOME /root
COPY ./paddle/scripts/docker/root/ /root/
RUN apt-get update && \
apt-get install -y --allow-downgrades \
git python-pip python-dev openssh-server bison \
apt-get install -y --allow-downgrades patchelf \
git python-pip python-dev python-opencv openssh-server bison \
libnccl2=2.1.2-1+cuda8.0 libnccl-dev=2.1.2-1+cuda8.0 \
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 \
automake locales clang-format swig doxygen cmake \
automake locales clang-format swig cmake \
liblapack-dev liblapacke-dev \
clang-3.8 llvm-3.8 libclang-3.8-dev \
net-tools libtool ccache && \
......@@ -70,7 +70,7 @@ RUN localedef -i en_US -f UTF-8 en_US.UTF-8
# specify sphinx version as 1.5.6 and remove -U option for [pip install -U
# sphinx-rtd-theme] since -U option will cause sphinx being updated to newest
# version(1.7.1 for now), which causes building documentation failed.
RUN pip install --upgrade pip==9.0.3 && \
RUN easy_install -U pip && \
pip install -U wheel && \
pip install -U docopt PyYAML sphinx==1.5.6 && \
pip install sphinx-rtd-theme==0.1.9 recommonmark
......@@ -79,6 +79,9 @@ RUN pip install pre-commit 'ipython==5.3.0' && \
pip install 'ipykernel==4.6.0' 'jupyter==1.0.0' && \
pip install opencv-python
#For docstring checker
RUN pip install pylint pytest astroid isort LinkChecker
COPY ./python/requirements.txt /root/
RUN pip install -r /root/requirements.txt
......@@ -101,6 +104,3 @@ RUN echo 'root:root' | chpasswd
RUN sed -ri 's/^PermitRootLogin\s+.*/PermitRootLogin yes/' /etc/ssh/sshd_config
RUN sed -ri 's/UsePAM yes/#UsePAM yes/g' /etc/ssh/sshd_config
EXPOSE 22
# development image default do build work
CMD ["bash", "/paddle/paddle/scripts/docker/build.sh"]
Dockerfile.android
浏览文件 @ a38a8db9
......@@ -40,5 +40,3 @@ RUN mkdir -p ${ANDROID_TOOLCHAINS_DIR} && \
unzip -q android-ndk-r14b-linux-x86_64.zip && \
mv android-ndk-r14b ${ANDROID_NDK_HOME} && \
rm -rf /opt/android-ndk-tmp
CMD ["bash", "/paddle/paddle/scripts/docker/build_android.sh"]
README.md
浏览文件 @ a38a8db9
......@@ -4,7 +4,6 @@
[![Build Status](https://travis-ci.org/PaddlePaddle/Paddle.svg?branch=develop)](https://travis-ci.org/PaddlePaddle/Paddle)
[![Documentation Status](https://img.shields.io/badge/docs-latest-brightgreen.svg?style=flat)](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/index_en.html)
[![Documentation Status](https://img.shields.io/badge/中文文档-最新-brightgreen.svg)](http://www.paddlepaddle.org/docs/develop/documentation/zh/getstarted/index_cn.html)
[![Coverage Status](https://coveralls.io/repos/github/PaddlePaddle/Paddle/badge.svg?branch=develop)](https://coveralls.io/github/PaddlePaddle/Paddle?branch=develop)
[![Release](https://img.shields.io/github/release/PaddlePaddle/Paddle.svg)](https://github.com/PaddlePaddle/Paddle/releases)
[![License](https://img.shields.io/badge/license-Apache%202-blue.svg)](LICENSE)
......@@ -19,6 +18,22 @@ learning to many products at Baidu.
Our vision is to enable deep learning for everyone via PaddlePaddle.
Please refer to our [release announcement](https://github.com/PaddlePaddle/Paddle/releases) to track the latest feature of PaddlePaddle.
### Latest PaddlePaddle Release: [Fluid 0.14.0](https://github.com/PaddlePaddle/Paddle/tree/v0.14.0)
### Install Latest Stable Release:
```
# Linux CPU
pip install paddlepaddle
# Linux GPU cuda9cudnn7
pip install paddlepaddle-gpu
# Linux GPU cuda8cudnn7
pip install paddlepaddle-gpu==0.14.0.post87
# Linux GPU cuda8cudnn5
pip install paddlepaddle-gpu==0.14.0.post85
# For installation on other platform, refer to http://paddlepaddle.org/
```
## Features
- **Flexibility**
......@@ -62,9 +77,9 @@ Please refer to our [release announcement](https://github.com/PaddlePaddle/Paddl
## Installation
It is recommended to check out the
[Docker installation guide](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/docker_install_en.html)
[Docker installation guide](http://www.paddlepaddle.org/docs/develop/documentation/fluid/en/build_and_install/docker_install_en.html)
before looking into the
[build from source guide](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/build_from_source_en.html).
[build from source guide](http://www.paddlepaddle.org/docs/develop/documentation/fluid/en/build_and_install/build_from_source_en.html).
## Documentation
......
benchmark/.gitignore
浏览文件 @ a38a8db9
......@@ -7,3 +7,6 @@ paddle/rnn/imdb.pkl
caffe/image/logs
tensorflow/image/logs
tensorflow/rnn/logs
fluid/models/*.pyc
fluid/logs
fluid/nohup.out
benchmark/cluster/README.md 已删除 100644 → 0
浏览文件 @ 10ec329b
# Cluster Training Benchmark
## Setup
- Platform
- Kubernetes: v1.6.2
- Linux Kernel: v3.10.0
- Resource
- CPU: 10 Cores per Pod
- Memory: 5GB per Pod
- Docker Image
We use different base Docker Image to run the benchmark on Kubernetes:
- PaddlePaddle v2: paddlepaddle/paddle:0.11.0
- PaddlePaddle Fluid: paddlepaddle/paddle:[commit-id]
- TensorFlow: tensorflow/tensorflow:1.5.0-rc0
- Model
vgg16 is used in this benchmark.
## Cases
- Variable
- Batch Size of training data.
- PServer count of the training job.
- The number of trainers.
- Invariant
- The resource of trainer/pserver Pod.
### Measure the Performance for Different Batch Size
- PServer Count: 40
- Trainer Count: 100
- Metrics: mini-batch / sec
<table>
<thead>
<tr>
<th>Batch Size </th>
<th> 32</th>
<th>64</th>
<th>128 </th>
<th>256</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>TensorFlow </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
</tbody>
</table>
### Measure the Performance for Different PServer Count
- Trainer Count: 100
- Batch Size: 64
- Metrics: mini-batch / sec
<table>
<thead>
<tr>
<th>PServer Count </th>
<th>10</th>
<th>20</th>
<th>40 </th>
<th>60</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>TensorFlow </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
</tr>
</tbody>
</table>
### Measure Parallel Efficiency By Increasing Trainer Count
- PServer Count: 20
- Batch Size: 64
- Metrics:
$S = \div(T1, TN)$
which S is the ratio of T1 over TN, training time of 1 and N trainers.
The parallel efficiency is:
$E = \div(S, N)$
<table>
<thead>
<tr>
<th>Trainer Counter </th>
<th>1</th>
<th>10</th>
<th>20 </th>
<th>30</th>
<th>40</th>
<th>50</th>
<th>60 </th>
<th>70</th>
<th>80</th>
<th>90</th>
<th>100 </th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
</tr>
<tr>
<td>TensorFlow </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
<td>- </td>
<td>-</td>
<td>- </td>
<td>- </td>
</tr>
</tbody>
</table>
## Reproduce the benchmark
TODO
benchmark/cluster/vgg16/Dockerfile 已删除 100644 → 0
浏览文件 @ 10ec329b
FROM nvidia/cuda:8.0-cudnn5-runtime-ubuntu16.04
# you can get mirror list here:
# https://launchpad.net/ubuntu/+archivemirrors
ARG UBUNTU_MIRROR
RUN /bin/bash -c 'if [[ -n ${UBUNTU_MIRROR} ]]; then sed -i 's#http://archive.ubuntu.com/ubuntu#${UBUNTU_MIRROR}#g' /etc/apt/sources.list; fi'
RUN apt-get update && apt-get install -y python python-dev python-pip iputils-ping libgtk2.0-dev
RUN pip install -U kubernetes opencv-python
RUN pip install paddlepaddle
# if network is slowly, you may need to add proxy here.
# ENV https_proxy=
RUN sh -c 'echo "import paddle.v2 as paddle\npaddle.dataset.cifar.train10()" | python'
RUN pip uninstall -y paddlepaddle
# unset proxy if it is setted.
# ENV https_proxy=""
# NOTE: By default CI built wheel packages turn WITH_DISTRIBUTE=OFF,
# so we must build one with distribute support to install in this image.
ADD *.whl /
RUN pip install /*.whl && rm -f /*.whl
ENV LD_LIBRARY_PATH=/usr/local/lib
# tf k8s
RUN pip install tensorflow==1.4.0
ADD tf_k8s /usr/bin
RUN chmod +x /usr/bin/tf_k8s
ADD vgg16_tf.py /workspace/
# below lines may change a lot for debugging
ADD https://raw.githubusercontent.com/PaddlePaddle/cloud/develop/docker/paddle_k8s /usr/bin
ADD https://raw.githubusercontent.com/PaddlePaddle/cloud/develop/docker/k8s_tools.py /root
RUN chmod +x /usr/bin/paddle_k8s
ADD vgg16_fluid.py vgg16_v2.py /workspace/
benchmark/cluster/vgg16/README.md 已删除 100644 → 0
浏览文件 @ 10ec329b
# Performance for Distributed vgg16
## Test Result
### Hardware Infomation
- CPU: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
- cpu MHz : 2101.000
- cache size : 20480 KB
### Blas settings
Setting environment variable: `MKL_NUM_THREADS=1`.
### Single Node Single Thread
- Metrics: samples / sec
<table>
<thead>
<tr>
<th>Batch Size </th>
<th> 32</th>
<th>64</th>
<th>128 </th>
<th>256</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td> 15.44 </td>
<td> 16.32 </td>
<td> 16.74 </td>
<td> 16.79 </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td> 15.97 </td>
<td> 17.04 </td>
<td> 17.60 </td>
<td> 17.83 </td>
</tr>
<tr>
<td>TensorFlow </td>
<td> 9.09 </td>
<td> 9.10 </td>
<td> 9.24 </td>
<td> 8.66 </td>
</tr>
</tbody>
</table>
### Different Batch Size
- PServer Count: 10
- Trainer Count: 20
- Metrics: samples / sec
<table>
<thead>
<tr>
<th>Batch Size </th>
<th> 32</th>
<th>64</th>
<th>128 </th>
<th>256</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td> 190.20 </td>
<td> 222.15 </td>
<td> 247.40 </td>
<td> 258.18 </td>
</tr>
<tr>
<td>PaddlePaddle v2 </td>
<td> 170.96 </td>
<td> 233.71 </td>
<td> 256.14 </td>
<td> 329.23 </td>
</tr>
<tr>
<td>TensorFlow </td>
<td> - </td>
<td> - </td>
<td> - </td>
<td> - </td>
</tr>
</tbody>
</table>
### Accelerate Rate
- Pserver Count: 20
- Batch Size: 128
- Metrics: samples / sec
<table>
<thead>
<tr>
<th>Trainer Count </th>
<th>20</th>
<th>40</th>
<th>80</th>
<th>100</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid</td>
<td> 263.29 (78.64%) </td>
<td> 518.80 (77.47%) </td>
<td> 836.26 (62.44%) </td>
<td> 1019.29 (60.89%) </td>
</tr>
<tr>
<td>PaddlePaddle v2 (need more tests) </td>
<td> 326.85 (92.85%) </td>
<td> 534.58 (75.93%) </td>
<td> 853.30 (60.60%) </td>
<td> 1041.99 (59.20%) </td>
</tr>
<tr>
<td>TensorFlow </td>
<td> - </td>
<td> - </td>
<td> - </td>
<td> - </td>
</tr>
</tbody>
</table>
### Different Pserver Count
- Trainer Count: 60
- Batch Size: 128
- Metrics: samples/ sec
<table>
<thead>
<tr>
<th>PServer Count </th>
<th>3</th>
<th>6</th>
<th>10</th>
<th>20</th>
</tr>
</thead>
<tbody>
<tr>
<td> PaddlePaddle Fluid(should fix in next PR) </td>
<td> 589.1 </td>
<td> 592.6 </td>
<td> 656.4 </td>
<td> 655.8 </td>
</tr>
<tr>
<td>PaddlePaddle v2 (need more tests) </td>
<td> 593.4 </td>
<td> 791.3 </td>
<td> 729.7 </td>
<td> 821.7 </td>
</tr>
<tr>
<td>TensorFlow </td>
<td> - </td>
<td> - </td>
<td> - </td>
<td> - </td>
</tr>
</tbody>
</table>
*The performance gap between Fuild and v2 comes from the network interference.*
## Steps to Run the Performance Test
1. You must re-compile PaddlePaddle and enable `-DWITH_DISTRIBUTE` to build PaddlePaddle with distributed support.
1. When the build finishes, copy the output `whl` package located under `build/python/dist` to current directory.
1. Run `docker build -t [image:tag] .` to build the docker image and run `docker push [image:tag]` to push the image to reponsitory so kubernetes can find it.
1. Run `kubectl create -f pserver.yaml && kubectl create -f trainer.yaml` to start the job on your kubernetes cluster (you must configure the `kubectl` client before this step).
1. Run `kubectl get po` to get running pods, and run `kubectl logs [podID]` to fetch the pod log of pservers and trainers.
Check the logs for the distributed training progress and analyze the performance.
## Enable Verbos Logs
Edit `pserver.yaml` and `trainer.yaml` and add an environment variable `GLOG_v=3` and `GLOG_logtostderr=1` to see what happend in detail.
benchmark/cluster/vgg16/fluid_pserver.yaml 已删除 100644 → 0
浏览文件 @ 10ec329b
apiVersion: extensions/v1beta1
kind: ReplicaSet
metadata:
name: vgg16job-pserver
spec:
replicas: 10
template:
metadata:
labels:
paddle-job-pserver: vgg16job
spec:
hostNetwork: true
imagePullSecrets:
- name: job-registry-secret
containers:
- name: pserver
image: "registry.baidu.com/paddlepaddle/fluid_benchmark:vgg16"
imagePullPolicy: Always
ports:
- name: jobport-30236
containerPort: 30236
env:
- name: PADDLE_JOB_NAME
value: vgg16job
- name: MKL_NUM_THREADS
value: "1"
- name: TRAINING_ROLE
value: "PSERVER"
- name: TRAINERS
value: "20"
- name: PSERVERS
value: "10"
- name: TOPOLOGY
value: ""
- name: ENTRY
value: "MKL_NUM_THREADS=1 python /workspace/vgg16_fluid.py --local 0"
- name: TRAINER_PACKAGE
value: "/workspace"
- name: PADDLE_INIT_PORT
value: "30236"
- name: PADDLE_INIT_NICS
value: "xgbe0"
- name: PADDLE_INIT_TRAINER_COUNT
value: "1"
- name: PADDLE_INIT_PORTS_NUM
value: "1"
- name: PADDLE_INIT_PORTS_NUM_FOR_SPARSE
value: "1"
- name: PADDLE_INIT_NUM_GRADIENT_SERVERS
value: "20"
- name: PADDLE_INIT_NUM_PASSES
value: "1"
- name: PADDLE_INIT_USE_GPU
value: "0"
- name: LD_LIBRARY_PATH
value: "/usr/local/lib:/usr/local/nvidia/lib64"
- name: NAMESPACE
valueFrom:
fieldRef:
fieldPath: "metadata.namespace"
- name: POD_IP
valueFrom:
fieldRef:
fieldPath: "status.podIP"
command: ["paddle_k8s", "start_fluid"]
resources:
requests:
memory: 10Gi
cpu: 4
limits:
memory: 10Gi
cpu: 4
benchmark/cluster/vgg16/fluid_trainer.yaml 已删除 100644 → 0
浏览文件 @ 10ec329b
apiVersion: batch/v1
kind: Job
metadata:
name: vgg16job-trainer
spec:
parallelism: 20
completions: 20
template:
metadata:
labels:
paddle-job: vgg16job
spec:
imagePullSecrets:
- name: job-registry-secret
hostNetwork: true
containers:
- name: trainer
image: "registry.baidu.com/paddlepaddle/fluid_benchmark:vgg16"
imagePullPolicy: Always
command: ["paddle_k8s", "start_fluid"]
env:
- name: PADDLE_JOB_NAME
value: vgg16job
- name: TRAINING_ROLE
value: "TRAINER"
- name: TRAINERS
value: "20"
- name: PSERVERS
value: "10"
- name: TOPOLOGY
value: ""
- name: ENTRY
value: "MKL_NUM_THREADS=1 python /workspace/vgg16_fluid.py --local 0 --batch_size 128"
- name: TRAINER_PACKAGE
value: "/workspace"
- name: PADDLE_INIT_PORT
value: "30236"
- name: PADDLE_INIT_NICS
value: "xgbe0"
- name: PADDLE_INIT_TRAINER_COUNT
value: "1"
- name: PADDLE_INIT_PORTS_NUM
value: "1"
- name: PADDLE_INIT_PORTS_NUM_FOR_SPARSE
value: "1"
- name: PADDLE_INIT_NUM_GRADIENT_SERVERS
value: "20"
- name: PADDLE_INIT_NUM_PASSES
value: "1"
- name: PADDLE_INIT_USE_GPU
value: "0"
- name: LD_LIBRARY_PATH
value: "/usr/local/lib:/usr/local/nvidia/lib64"
- name: NAMESPACE
valueFrom:
fieldRef:
fieldPath: "metadata.namespace"
- name: POD_IP
valueFrom:
fieldRef:
fieldPath: "status.podIP"
resources:
requests:
memory: 40Gi
cpu: 2
limits:
memory: 40Gi
cpu: 2
restartPolicy: Never
benchmark/cluster/vgg16/run_vgg_dist.sh 已删除 100644 → 0
浏览文件 @ 10ec329b
#!/bin/bash
# Update to point to the source file.
VGG_SRC="vgg16_fluid.py"
export TRAINING_ROLE=PSERVER
export TRAINERS=2
export POD_IP=127.0.0.1
export PADDLE_INIT_PORT=6174
MKL_NUM_THREADS=1 python -u ${VGG_SRC} --local 0 --ps_host=127.0.0.1:6174 --trainer_hosts=127.0.0.1:6174 &
# Need to wait for the ps to start first.
sleep 10
echo "done start ps"
export TRAINING_ROLE=TRAINER
export TRAINERS=2
export POD_IP=127.0.0.1
export PADDLE_INIT_PORT=6174
CUDA_VISIBLE_DEVICES=4 MKL_NUM_THREADS=1 python -u ${VGG_SRC} --local 0 --ps_host=127.0.0.1:6174 --trainer_hosts=127.0.0.1:6174 --device=GPU --task_index=0 &
CUDA_VISIBLE_DEVICES=5 MKL_NUM_THREADS=1 python -u ${VGG_SRC} --local 0 --ps_host=127.0.0.1:6174 --trainer_hosts=127.0.0.1:6174 --device=GPU --task_index=1 &
benchmark/cluster/vgg16/tf_k8s 已删除 100644 → 0
浏览文件 @ 10ec329b
#!/bin/bash
check_trainer_ret() {
ret=$1
stdbuf -oL echo "job returned $ret...setting pod return message..."
stdbuf -oL echo "==============================="
if [ $ret -eq 136 ] ; then
echo "Error Arithmetic Operation(Floating Point Exception)" > /dev/termination-log
elif [ $ret -eq 139 ] ; then
echo "Segmentation Fault" > /dev/termination-log
elif [ $ret -eq 1 ] ; then
echo "General Error" > /dev/termination-log
elif [ $ret -eq 134 ] ; then
echo "Program Abort" > /dev/termination-log
fi
stdbuf -oL echo "termination log wroted..."
exit $ret
}
g_pservers=""
g_trainers=""
wait_running_pods(){
pserver_label="tf-job-pserver=${JOB_NAME}"
trainer_label="tf-job-trainer=${JOB_NAME}"
stdbuf -oL python /root/k8s_tools.py wait_pods_running ${pserver_label} ${PSERVERS_NUM}
stdbuf -oL python /root/k8s_tools.py wait_pods_running ${trainer_label} ${TRAINERS_NUM}
g_pservers=$(python /root/k8s_tools.py fetch_endpoints ${pserver_label} ${PORT})
g_trainers=$(python /root/k8s_tools.py fetch_endpoints ${trainer_label} ${PORT})
}
start_tf_pserver(){
wait_running_pods
label="tf-job-pserver=${JOB_NAME}"
pserver_id=$(python /root/k8s_tools.py fetch_id ${label})
cmd="${ENTRY} --ps_hosts=${g_pservers} --worker_hosts=${g_trainers} \
--job_name=${TF_JOB_NAME} --task_index=${pserver_id}"
stdbuf -oL sh -c "cd ${TRAINER_PACKAGE} && ${cmd}"
}
start_tf_trainer(){
wait_running_pods
label="tf-job-trainer=${JOB_NAME}"
trainer_id=$(python /root/k8s_tools.py fetch_id ${label})
cmd="${ENTRY} --ps_hosts=${g_pservers} --worker_hosts=${g_trainers} \
--job_name=${TF_JOB_NAME} --task_index=${trainer_id} --batch_size=${BATCH_SIZE}"
stdbuf -oL sh -c "cd ${TRAINER_PACKAGE} && ${cmd}"
check_trainer_ret $?
}
start_tf(){
if [[ "${TF_JOB_NAME}" == "worker" ]]; then
start_tf_trainer
else
start_tf_pserver
fi
}
usage() {
echo "usage: tf_k8s [<args>]:"
echo " start_tf Start tensorflow jobs"
}
case "$1" in
start_tf)
start_tf
;;
--help)
usage
;;
*)
usage
;;
esac
benchmark/cluster/vgg16/tf_pserver.yaml 已删除 100644 → 0
浏览文件 @ 10ec329b
apiVersion: extensions/v1beta1
kind: ReplicaSet
metadata:
name: vgg16job-tf-pserver
spec:
replicas: 10
template:
metadata:
labels:
tf-job-pserver: vgg16job-tf
spec:
hostNetwork: true
imagePullSecrets:
- name: job-registry-secret
containers:
- name: pserver
image: "registry.baidu.com/paddlepaddle/fluid_benchmark_tf:vgg16"
imagePullPolicy: Always
command: ["tf_k8s", "start_tf"]
ports:
- name: jobport-30236
containerPort: 30236
env:
- name: PORT
value: "32036"
- name: ENTRY
value: "python vgg16_tf.py"
- name: JOB_NAME
value: vgg16job-tf
- name: PSERVERS_NUM
value: "10"
- name: TF_JOB_NAME
value: "ps"
- name: TRAINERS_NUM
value: "20"
- name: BATCH_SIZE
value: "128"
- name: TRAINER_PACKAGE
value: "/workspace"
- name: NUM_PASSES
value: "1"
- name: NAMESPACE
valueFrom:
fieldRef:
fieldPath: "metadata.namespace"
- name: POD_IP
valueFrom:
fieldRef:
fieldPath: "status.podIP"
resources:
requests:
memory: 10Gi
cpu: 4
limits:
memory: 10Gi
cpu: 4
benchmark/cluster/vgg16/tf_trainer.yaml 已删除 100644 → 0
浏览文件 @ 10ec329b
apiVersion: batch/v1
kind: Job
metadata:
name: vgg16job-tf-trainer
spec:
parallelism: 20
completions: 20
template:
metadata:
labels:
tf-job-trainer: vgg16job-tf
spec:
imagePullSecrets:
- name: job-registry-secret
hostNetwork: true
containers:
- name: trainer
image: "registry.baidu.com/paddlepaddle/fluid_benchmark_tf:vgg16"
imagePullPolicy: Always
command: ["tf_k8s", "start_tf"]
ports:
- name: jobport-30236
containerPort: 30236
env:
- name: PORT
value: "32036"
- name: JOB_NAME
value: vgg16job-tf
- name: TF_JOB_NAME
value: "worker"
- name: ENTRY
value: "python vgg16_tf.py"
- name: PSERVERS_NUM
value: "10"
- name: BATCH_SIZE
value: "128"
- name: TRAINERS_NUM
value: "20"
- name: TRAINER_PACKAGE
value: "/workspace"
- name: NUM_PASSES
value: "1"
- name: NAMESPACE
valueFrom:
fieldRef:
fieldPath: "metadata.namespace"
- name: POD_IP
valueFrom:
fieldRef:
fieldPath: "status.podIP"
resources:
requests:
memory: 40Gi
cpu: 2
limits:
memory: 40Gi
cpu: 2
restartPolicy: Never
benchmark/cluster/vgg16/v2_pserver.yaml 已删除 100644 → 0
浏览文件 @ 10ec329b
apiVersion: extensions/v1beta1
kind: ReplicaSet
metadata:
name: vgg16v2job-pserver
spec:
replicas: 10
template:
metadata:
labels:
paddle-job-pserver: vgg16v2job
spec:
hostNetwork: true
imagePullSecrets:
- name: job-registry-secret
containers:
- name: pserver
image: "registry.baidu.com/paddlepaddle/fluid_benchmark:vgg16"
imagePullPolicy: Always
ports:
- name: jobport-30236
containerPort: 30236
env:
- name: PADDLE_JOB_NAME
value: vgg16v2job
- name: TRAINERS
value: "20"
- name: PSERVERS
value: "10"
- name: TOPOLOGY
value: ""
- name: ENTRY
value: "python train.py"
- name: TRAINER_PACKAGE
value: "/workspace"
- name: PADDLE_INIT_PORT
value: "30236"
- name: PADDLE_INIT_NICS
value: "xgbe0"
- name: PADDLE_INIT_TRAINER_COUNT
value: "1"
- name: PADDLE_INIT_PORTS_NUM
value: "1"
- name: PADDLE_INIT_PORTS_NUM_FOR_SPARSE
value: "1"
- name: PADDLE_INIT_NUM_GRADIENT_SERVERS
value: "20"
- name: PADDLE_INIT_NUM_PASSES
value: "1"
- name: PADDLE_INIT_USE_GPU
value: "0"
- name: LD_LIBRARY_PATH
value: "/usr/local/lib:/usr/local/nvidia/lib64"
- name: NAMESPACE
valueFrom:
fieldRef:
fieldPath: "metadata.namespace"
command: ["paddle_k8s", "start_pserver"]
resources:
requests:
memory: 10Gi
cpu: 4
limits:
memory: 10Gi
cpu: 4
benchmark/cluster/vgg16/v2_trainer.yaml 已删除 100644 → 0
浏览文件 @ 10ec329b
apiVersion: batch/v1
kind: Job
metadata:
name: vgg16v2job-trainer
spec:
parallelism: 20
completions: 20
template:
metadata:
labels:
paddle-job: vgg16v2job
spec:
imagePullSecrets:
- name: job-registry-secret
hostNetwork: true
containers:
- name: trainer
image: "registry.baidu.com/paddlepaddle/fluid_benchmark:vgg16"
imagePullPolicy: Always
command: ["paddle_k8s", "start_trainer", "v2"]
env:
- name: PADDLE_JOB_NAME
value: vgg16v2job
- name: BATCH_SIZE
value: "256"
- name: TRAINERS
value: "20"
- name: PSERVERS
value: "10"
- name: TOPOLOGY
value: ""
- name: ENTRY
value: "cd /workspace && MKL_NUM_THREADS=1 python /workspace/vgg16_v2.py"
- name: TRAINER_PACKAGE
value: "/workspace"
- name: PADDLE_INIT_PORT
value: "30236"
- name: PADDLE_INIT_NICS
value: "xgbe0"
- name: PADDLE_INIT_TRAINER_COUNT
value: "1"
- name: PADDLE_INIT_PORTS_NUM
value: "1"
- name: PADDLE_INIT_PORTS_NUM_FOR_SPARSE
value: "1"
- name: PADDLE_INIT_NUM_GRADIENT_SERVERS
value: "20"
- name: PADDLE_INIT_NUM_PASSES
value: "2"
- name: PADDLE_INIT_USE_GPU
value: "0"
- name: LD_LIBRARY_PATH
value: "/usr/local/lib:/usr/local/nvidia/lib64"
- name: NAMESPACE
valueFrom:
fieldRef:
fieldPath: "metadata.namespace"
resources:
requests:
memory: 40Gi
cpu: 2
limits:
memory: 40Gi
cpu: 2
restartPolicy: Never
benchmark/cluster/vgg16/vgg16_fluid.py 已删除 100644 → 0
浏览文件 @ 10ec329b
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""VGG16 benchmark in Fluid"""
from __future__ import print_function
import sys
import time
import numpy as np
import paddle.v2 as paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.profiler as profiler
import argparse
import functools
import os
from paddle.fluid import debuger
def str2bool(v):
if v.lower() in ('yes', 'true', 't', 'y', '1'):
return True
elif v.lower() in ('no', 'false', 'f', 'n', '0'):
return False
else:
raise argparse.ArgumentTypeError('Boolean value expected.')
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
'--batch_size', type=int, default=128, help="Batch size for training.")
parser.add_argument(
'--learning_rate',
type=float,
default=1e-3,
help="Learning rate for training.")
parser.add_argument('--num_passes', type=int, default=50, help="No. of passes.")
parser.add_argument(
'--device',
type=str,
default='CPU',
choices=['CPU', 'GPU'],
help="The device type.")
parser.add_argument('--device_id', type=int, default=0, help="The device id.")
parser.add_argument(
'--data_format',
type=str,
default='NCHW',
choices=['NCHW', 'NHWC'],
help='The data order, now only support NCHW.')
parser.add_argument(
'--data_set',
type=str,
default='cifar10',
choices=['cifar10', 'flowers'],
help='Optional dataset for benchmark.')
parser.add_argument(
'--local',
type=str2bool,
default=True,
help='Whether to run as local mode.')
parser.add_argument(
"--ps_hosts",
type=str,
default="",
help="Comma-separated list of hostname:port pairs")
parser.add_argument(
"--trainer_hosts",
type=str,
default="",
help="Comma-separated list of hostname:port pairs")
parser.add_argument(
"--profile", action='store_true', help="If set, profile a few steps.")
# Flags for defining the tf.train.Server
parser.add_argument(
"--task_index", type=int, default=0, help="Index of task within the job")
args = parser.parse_args()
def vgg16_bn_drop(input):
def conv_block(input, num_filter, groups, dropouts):
return fluid.nets.img_conv_group(
input=input,
pool_size=2,
pool_stride=2,
conv_num_filter=[num_filter] * groups,
conv_filter_size=3,
conv_act='relu',
conv_with_batchnorm=True,
conv_batchnorm_drop_rate=dropouts,
pool_type='max')
conv1 = conv_block(input, 64, 2, [0.3, 0])
conv2 = conv_block(conv1, 128, 2, [0.4, 0])
conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0])
conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0])
conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0])
drop = fluid.layers.dropout(x=conv5, dropout_prob=0.5)
fc1 = fluid.layers.fc(input=drop, size=4096, act=None)
bn = fluid.layers.batch_norm(input=fc1, act='relu')
drop2 = fluid.layers.dropout(x=bn, dropout_prob=0.5)
fc2 = fluid.layers.fc(input=drop2, size=4096, act=None)
return fc2
def main():
if args.data_set == "cifar10":
classdim = 10
if args.data_format == 'NCHW':
data_shape = [3, 32, 32]
else:
data_shape = [32, 32, 3]
else:
classdim = 102
if args.data_format == 'NCHW':
data_shape = [3, 224, 224]
else:
data_shape = [224, 224, 3]
# Input data
images = fluid.layers.data(name='pixel', shape=data_shape, dtype='float32')
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
# Train program
net = vgg16_bn_drop(images)
predict = fluid.layers.fc(input=net, size=classdim, act='softmax')
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
# Evaluator
batch_size = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(
input=predict, label=label, total=batch_size)
# inference program
inference_program = fluid.default_main_program().clone()
with fluid.program_guard(inference_program):
inference_program = fluid.io.get_inference_program(batch_acc)
# Optimization
optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
optimize_ops, params_grads = optimizer.minimize(avg_cost)
# Initialize executor
place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(
args.device_id)
exe = fluid.Executor(place)
# test
def test(exe):
test_pass_acc = fluid.average.WeightedAverage()
for batch_id, data in enumerate(test_reader()):
img_data = np.array(map(lambda x: x[0].reshape(data_shape),
data)).astype("float32")
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([-1, 1])
outs = exe.run(inference_program,
feed={"pixel": img_data,
"label": y_data},
fetch_list=[batch_acc, batch_size])
test_pass_acc.add(value=np.array(outs[0]), weight=np.array(outs[1]))
return test_pass_acc.eval()
def train_loop(exe, trainer_prog):
iters = 0
ts = time.time()
train_pass_acc = fluid.average.WeightedAverage()
for pass_id in range(args.num_passes):
# train
start_time = time.time()
num_samples = 0
train_pass_acc.reset()
def run_step(batch_id, data):
img_data = np.array(
map(lambda x: x[0].reshape(data_shape), data)).astype(
"float32")
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([-1, 1])
loss, acc, b_size = exe.run(
trainer_prog,
feed={"pixel": img_data,
"label": y_data},
fetch_list=[avg_cost, batch_acc, batch_size])
return loss, acc, b_size
if args.profile and args.task_index == 0:
# warmup.
for batch_id, data in enumerate(train_reader()):
if batch_id > 5: break
run_step(batch_id, data)
with profiler.profiler('All', 'total', '/tmp/profile_vgg'):
for batch_id, data in enumerate(train_reader()):
if batch_id > 5: break
run_step(batch_id, data)
for batch_id, data in enumerate(train_reader()):
ts = time.time()
loss, acc, b_size = run_step(batch_id, data)
iters += 1
num_samples += len(data)
train_pass_acc.add(value=acc, weight=b_size)
print(
"Pass = %d, Iters = %d, Loss = %f, Accuracy = %f, "
"Speed = %.2f img/s" % (pass_id, iters, loss, acc,
len(data) / (time.time() - ts))
) # The accuracy is the accumulation of batches, but not the current batch.
pass_elapsed = time.time() - start_time
pass_train_acc = train_pass_acc.eval()
pass_test_acc = test(exe)
print("Task:%d Pass = %d, Training performance = %f imgs/s, "
"Train accuracy = %f, Test accuracy = %f\n" %
(args.task_index, pass_id, num_samples / pass_elapsed,
pass_train_acc, pass_test_acc))
if args.local:
# Parameter initialization
exe.run(fluid.default_startup_program())
# data reader
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.train10() if args.data_set == 'cifar10'
else paddle.dataset.flowers.train(),
buf_size=5120),
batch_size=args.batch_size)
test_reader = paddle.batch(
paddle.dataset.cifar.test10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.test(),
batch_size=args.batch_size)
train_loop(exe, fluid.default_main_program())
else:
trainers = int(os.getenv("TRAINERS")) # total trainer count
print("trainers total: ", trainers)
training_role = os.getenv(
"TRAINING_ROLE",
"TRAINER") # get the training role: trainer/pserver
t = fluid.DistributeTranspiler()
t.transpile(
trainer_id=args.task_index,
pservers=args.ps_hosts,
trainers=trainers)
if training_role == "PSERVER":
current_endpoint = os.getenv("POD_IP") + ":" + os.getenv(
"PADDLE_INIT_PORT")
if not current_endpoint:
print("need env SERVER_ENDPOINT")
exit(1)
pserver_prog = t.get_pserver_program(current_endpoint)
pserver_startup = t.get_startup_program(current_endpoint,
pserver_prog)
exe.run(pserver_startup)
exe.run(pserver_prog)
elif training_role == "TRAINER":
# Parameter initialization
exe.run(fluid.default_startup_program())
# data reader
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.train10() if args.data_set == 'cifar10'
else paddle.dataset.flowers.train(),
buf_size=5120),
batch_size=args.batch_size)
test_reader = paddle.batch(
paddle.dataset.cifar.test10() if args.data_set == 'cifar10' else
paddle.dataset.flowers.test(),
batch_size=args.batch_size)
trainer_prog = t.get_trainer_program()
feeder = fluid.DataFeeder(feed_list=[images, label], place=place)
# TODO(typhoonzero): change trainer startup program to fetch parameters from pserver
exe.run(fluid.default_startup_program())
train_loop(exe, trainer_prog)
else:
print("environment var TRAINER_ROLE should be TRAINER os PSERVER")
def print_arguments():
print('----------- Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
if __name__ == "__main__":
print_arguments()
main()
benchmark/cluster/vgg16/vgg16_tf.py 已删除 100644 → 0
浏览文件 @ 10ec329b
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""VGG16 benchmark in TensorFlow
You can get distribution example template structure here:
https://medium.com/clusterone/how-to-write-distributed-tensorflow-code-with-an-example-on-tensorport-70bf3306adcb
https://www.tensorflow.org/deploy/distributed
"""
import tensorflow as tf
import paddle.v2 as paddle
import numpy as np
import argparse
import time
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
'--batch_size', type=int, default=128, help="Batch size for training.")
parser.add_argument(
'--learning_rate',
type=float,
default=1e-3,
help="Learning rate for training.")
parser.add_argument('--num_passes', type=int, default=50, help="No. of passes.")
parser.add_argument(
'--device',
type=str,
default='CPU',
choices=['CPU', 'GPU'],
help="The device type.")
parser.add_argument(
'--data_format',
type=str,
default='NHWC',
choices=['NCHW', 'NHWC'],
help='The data order, NCHW=[batch, channels, height, width].'
'Only support NHWC right now.')
parser.add_argument(
'--data_set',
type=str,
default='cifar10',
choices=['cifar10', 'flowers'],
help='Optional dataset for benchmark.')
parser.add_argument(
"--ps_hosts",
type=str,
default="",
help="Comma-separated list of hostname:port pairs")
parser.add_argument(
"--worker_hosts",
type=str,
default="",
help="Comma-separated list of hostname:port pairs")
parser.add_argument(
"--job_name", type=str, default="", help="One of 'worker', 'ps'")
# Flags for defining the tf.train.Server
parser.add_argument(
"--task_index", type=int, default=0, help="Index of task within the job")
args = parser.parse_args()
class VGG16Model(object):
def __init__(self):
self.parameters = []
def batch_norm_relu(self, inputs, is_training):
"""Performs a batch normalization followed by a ReLU."""
# We set fused=True for a significant speed boost. See
# https://www.tensorflow.org/speed/speed_guide#common_fused_ops
inputs = tf.layers.batch_normalization(
inputs=inputs,
axis=1 if args.data_format == 'NCHW' else -1,
momentum=0.9,
epsilon=1e-05,
center=True,
scale=True,
training=is_training,
fused=True)
inputs = tf.nn.relu(inputs)
return inputs
def conv_bn_layer(self,
name,
images,
kernel_shape,
is_training,
drop_rate=0.0):
with tf.name_scope(name) as scope:
kernel = tf.Variable(
tf.truncated_normal(
kernel_shape, dtype=tf.float32, stddev=1e-1),
name='weights')
conv = tf.nn.conv2d(
images,
kernel, [1, 1, 1, 1],
data_format=args.data_format,
padding='SAME')
biases = tf.Variable(
tf.constant(
0.0, shape=[kernel_shape[-1]], dtype=tf.float32),
trainable=True,
name='biases')
out = tf.nn.bias_add(conv, biases)
out = self.batch_norm_relu(out, is_training)
out = tf.layers.dropout(out, rate=drop_rate, training=is_training)
return out
def fc_layer(self, name, inputs, shape):
with tf.name_scope(name) as scope:
fc_w = tf.Variable(
tf.truncated_normal(
shape, dtype=tf.float32, stddev=1e-1),
name='weights')
fc_b = tf.Variable(
tf.constant(
0.0, shape=[shape[-1]], dtype=tf.float32),
trainable=True,
name='biases')
out = tf.nn.bias_add(tf.matmul(inputs, fc_w), fc_b)
return out
def network(self, images, class_dim, is_training):
""" VGG16 model structure.
TODO(kuke): enable this network to support the 'NCHW' data format
"""
# conv1
conv1_1 = self.conv_bn_layer(
'conv1_1', images, [3, 3, 3, 64], is_training, drop_rate=0.3)
conv1_2 = self.conv_bn_layer(
'conv1_2', conv1_1, [3, 3, 64, 64], is_training, drop_rate=0.0)
# pool1
pool1 = tf.nn.max_pool(
conv1_2,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool1')
# conv2
conv2_1 = self.conv_bn_layer(
'conv2_1', pool1, [3, 3, 64, 128], is_training, drop_rate=0.4)
conv2_2 = self.conv_bn_layer(
'conv2_2', conv2_1, [3, 3, 128, 128], is_training, drop_rate=0.0)
# pool2
pool2 = tf.nn.max_pool(
conv2_2,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool2')
# conv3
conv3_1 = self.conv_bn_layer(
'conv3_1', pool2, [3, 3, 128, 256], is_training, drop_rate=0.4)
conv3_2 = self.conv_bn_layer(
'conv3_2', conv3_1, [3, 3, 256, 256], is_training, drop_rate=0.4)
conv3_3 = self.conv_bn_layer(
'conv3_3', conv3_2, [3, 3, 256, 256], is_training, drop_rate=0.0)
# pool3
pool3 = tf.nn.max_pool(
conv3_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool3')
# conv4
conv4_1 = self.conv_bn_layer(
'conv4_1', pool3, [3, 3, 256, 512], is_training, drop_rate=0.4)
conv4_2 = self.conv_bn_layer(
'conv4_2', conv4_1, [3, 3, 512, 512], is_training, drop_rate=0.4)
conv4_3 = self.conv_bn_layer(
'conv4_3', conv4_2, [3, 3, 512, 512], is_training, drop_rate=0.0)
# pool4
pool4 = tf.nn.max_pool(
conv4_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool4')
# conv5
conv5_1 = self.conv_bn_layer(
'conv5_1', pool4, [3, 3, 512, 512], is_training, drop_rate=0.4)
conv5_2 = self.conv_bn_layer(
'conv5_2', conv5_1, [3, 3, 512, 512], is_training, drop_rate=0.4)
conv5_3 = self.conv_bn_layer(
'conv5_3', conv5_2, [3, 3, 512, 512], is_training, drop_rate=0.0)
# pool5
pool5 = tf.nn.max_pool(
conv5_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool4')
# flatten
shape = int(np.prod(pool5.get_shape()[1:]))
pool5_flat = tf.reshape(pool5, [-1, shape])
# fc1
drop = tf.layers.dropout(pool5_flat, rate=0.5, training=is_training)
fc1 = self.fc_layer('fc1', drop, [shape, 512])
# fc2
bn = self.batch_norm_relu(fc1, is_training)
drop = tf.layers.dropout(bn, rate=0.5, training=is_training)
fc2 = self.fc_layer('fc2', drop, [512, 512])
fc3 = self.fc_layer('fc3', fc2, [512, class_dim])
return fc3
def run_benchmark(cluster_spec, server):
"""Run benchmark on cifar10 or flowers."""
if args.data_set == "cifar10":
class_dim = 10
raw_shape = (3, 32, 32)
dat_shape = (None, 32, 32, 3) if args.data_format == 'NHWC' else (
None, 3, 32, 32)
else:
class_dim = 102
raw_shape = (3, 224, 224)
dat_shape = (None, 224, 224, 3) if args.data_format == 'NHWC' else (
None, 3, 224, 224)
device = tf.train.replica_device_setter(
worker_device="/job:worker/task:{}".format(args.task_index),
cluster=cluster_spec)
with tf.device(device):
images = tf.placeholder(tf.float32, shape=dat_shape)
labels = tf.placeholder(tf.int64, shape=(None, ))
is_training = tf.placeholder('bool')
onehot_labels = tf.one_hot(labels, depth=class_dim)
vgg16 = VGG16Model()
logits = vgg16.network(images, class_dim, is_training)
loss = tf.losses.softmax_cross_entropy(
onehot_labels=onehot_labels, logits=logits)
avg_loss = tf.reduce_mean(loss)
correct = tf.equal(tf.argmax(logits, 1), labels)
accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))
optimizer = tf.train.AdamOptimizer(learning_rate=args.learning_rate)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
global_step = tf.Variable(0, name='global_step', trainable=False)
with tf.control_dependencies(update_ops):
train_op = optimizer.minimize(avg_loss, global_step=global_step)
summary_op = tf.summary.merge_all()
init_op = tf.global_variables_initializer()
# data reader
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.train10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.train(),
buf_size=5120),
batch_size=args.batch_size)
test_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.test10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.test(),
buf_size=5120),
batch_size=args.batch_size)
# test
def test():
test_accs = []
for batch_id, data in enumerate(test_reader()):
test_images = np.array(
map(lambda x: np.transpose(x[0].reshape(raw_shape),
axes=[1, 2, 0]) if args.data_format == 'NHWC' else x[0], data)).astype("float32")
test_labels = np.array(map(lambda x: x[1], data)).astype('int64')
test_accs.append(
accuracy.eval(feed_dict={
images: test_images,
labels: test_labels,
is_training: False
}))
return np.mean(test_accs)
config = tf.ConfigProto(
intra_op_parallelism_threads=1,
inter_op_parallelism_threads=1,
log_device_placement=True)
config.gpu_options.allow_growth = True
hooks = [tf.train.StopAtStepHook(last_step=1000000)]
with tf.train.MonitoredTrainingSession(
master=server.target,
is_chief=(args.task_index == 0),
hooks=hooks,
config=config) as sess:
iters, num_samples, start_time = 0, 0, 0.0
for pass_id in range(args.num_passes):
# train
num_samples = 0
start_time = time.time()
for batch_id, data in enumerate(train_reader()):
train_images = np.array(
map(lambda x: np.transpose(x[0].reshape(raw_shape),
axes=[1, 2, 0]) if args.data_format == 'NHWC' else x[0], data)).astype("float32")
train_labels = np.array(map(lambda x: x[1], data)).astype(
'int64')
iter_begin_time = time.time()
_, loss, acc = sess.run([train_op, avg_loss, accuracy],
feed_dict={
images: train_images,
labels: train_labels,
is_training: True
})
iters += 1
print(
"Pass = %d, Iters = %d, Loss = %f, Accuracy = %f, Speed=%.2f imgs/sec"
% (pass_id, iters, loss, acc,
len(data) / (time.time() - iter_begin_time)))
num_samples += len(data)
train_elapsed = time.time() - start_time
# test
pass_test_acc = test()
print("Pass = %d, Train speed = %f imgs/s, Test accuracy = %f\n" %
(pass_id, num_samples / train_elapsed, pass_test_acc))
def print_arguments():
print('----------- Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
if __name__ == '__main__':
print_arguments()
ps_hosts = args.ps_hosts.split(",")
worker_hosts = args.worker_hosts.split(",")
# Create a cluster from the parameter server and worker hosts.
cluster_spec = tf.train.ClusterSpec({
"ps": ps_hosts,
"worker": worker_hosts
})
# Create and start a server for the local task.
server = tf.train.Server(
cluster_spec, job_name=args.job_name, task_index=args.task_index)
if args.job_name == "ps":
print("start pserver")
server.join()
elif args.job_name == "worker":
print("start worker")
run_benchmark(cluster_spec, server)
benchmark/cluster/vgg16/vgg16_v2.py 已删除 100644 → 0
浏览文件 @ 10ec329b
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
#Licensed under the Apache License, Version 2.0 (the "License");
#you may not use this file except in compliance with the License.
#You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
#Unless required by applicable law or agreed to in writing, software
#distributed under the License is distributed on an "AS IS" BASIS,
#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#See the License for the specific language governing permissions and
#limitations under the License.
import gzip
import paddle.v2.dataset.cifar as cifar
import paddle.v2 as paddle
import time
import os
DATA_DIM = 3 * 32 * 32
CLASS_DIM = 10
BATCH_SIZE = os.getenv("BATCH_SIZE")
if BATCH_SIZE:
BATCH_SIZE = int(BATCH_SIZE)
else:
BATCH_SIZE = 128
print "batch_size", BATCH_SIZE
NODE_COUNT = int(os.getenv("TRAINERS"))
ts = 0
def vgg(input, nums, class_dim):
def conv_block(input, num_filter, groups, num_channels=None):
return paddle.networks.img_conv_group(
input=input,
num_channels=num_channels,
pool_size=2,
pool_stride=2,
conv_num_filter=[num_filter] * groups,
conv_filter_size=3,
conv_act=paddle.activation.Relu(),
pool_type=paddle.pooling.Max())
assert len(nums) == 5
# the channel of input feature is 3
conv1 = conv_block(input, 64, nums[0], 3)
conv2 = conv_block(conv1, 128, nums[1])
conv3 = conv_block(conv2, 256, nums[2])
conv4 = conv_block(conv3, 512, nums[3])
conv5 = conv_block(conv4, 512, nums[4])
fc_dim = 512
fc1 = paddle.layer.fc(input=conv5,
size=fc_dim,
act=paddle.activation.Relu(),
layer_attr=paddle.attr.Extra(drop_rate=0.5))
fc2 = paddle.layer.fc(input=fc1,
size=fc_dim,
act=paddle.activation.Relu(),
layer_attr=paddle.attr.Extra(drop_rate=0.5))
out = paddle.layer.fc(input=fc2,
size=class_dim,
act=paddle.activation.Softmax())
return out
def vgg13(input, class_dim):
nums = [2, 2, 2, 2, 2]
return vgg(input, nums, class_dim)
def vgg16(input, class_dim):
nums = [2, 2, 3, 3, 3]
return vgg(input, nums, class_dim)
def vgg19(input, class_dim):
nums = [2, 2, 4, 4, 4]
return vgg(input, nums, class_dim)
def main():
global ts
paddle.init(use_gpu=False)
image = paddle.layer.data(
name="image", type=paddle.data_type.dense_vector(DATA_DIM))
lbl = paddle.layer.data(
name="label", type=paddle.data_type.integer_value(CLASS_DIM))
extra_layers = None
# NOTE: for v2 distributed training need averaging updates.
learning_rate = 1e-3 / NODE_COUNT
out = vgg16(image, class_dim=CLASS_DIM)
cost = paddle.layer.classification_cost(input=out, label=lbl)
# Create parameters
parameters = paddle.parameters.create(cost)
# Create optimizer
optimizer = paddle.optimizer.Momentum(
momentum=0.9,
regularization=paddle.optimizer.L2Regularization(rate=0.0005 *
BATCH_SIZE),
learning_rate=learning_rate / BATCH_SIZE,
learning_rate_decay_a=0.1,
learning_rate_decay_b=128000 * 35,
learning_rate_schedule="discexp", )
train_reader = paddle.batch(
paddle.reader.shuffle(
cifar.train10(),
# To use other data, replace the above line with:
# reader.train_reader('train.list'),
buf_size=1000),
batch_size=BATCH_SIZE)
test_reader = paddle.batch(
cifar.test10(),
# To use other data, replace the above line with:
# reader.test_reader('val.list'),
batch_size=BATCH_SIZE)
# Create trainer
trainer = paddle.trainer.SGD(cost=cost,
parameters=parameters,
update_equation=optimizer,
extra_layers=extra_layers,
is_local=False)
# End batch and end pass event handler
def event_handler(event):
global ts, ts_pass
if isinstance(event, paddle.event.BeginPass):
ts_pass = time.time()
if isinstance(event, paddle.event.BeginIteration):
ts = time.time()
if isinstance(event, paddle.event.EndIteration):
if event.batch_id % 1 == 0:
print "\nPass %d, Batch %d, Cost %f, %s, spent: %f" % (
event.pass_id, event.batch_id, event.cost, event.metrics,
time.time() - ts)
if isinstance(event, paddle.event.EndPass):
print "Pass %d end, spent: %f" % (event.pass_id,
time.time() - ts_pass)
result = trainer.test(reader=test_reader)
print "\nTest with Pass %d, %s" % (event.pass_id, result.metrics)
trainer.train(
reader=train_reader, num_passes=200, event_handler=event_handler)
if __name__ == '__main__':
main()
benchmark/fluid/Dockerfile 0 → 100644
浏览文件 @ a38a8db9
FROM nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04
# Use UBUNTU_MIRROR can speed up apt-get speed.
# ARG UBUNTU_MIRROR
# RUN /bin/bash -c 'if [[ -n ${UBUNTU_MIRROR} ]]; then sed -i 's#http://archive.ubuntu.com/ubuntu#${UBUNTU_MIRROR}#g' /etc/apt/sources.list; fi'
RUN apt-get update && apt-get install -y python python-pip iputils-ping libgtk2.0-dev wget vim net-tools iftop python-opencv
RUN ln -s /usr/lib/x86_64-linux-gnu/libcudnn.so.7 /usr/lib/libcudnn.so && ln -s /usr/lib/x86_64-linux-gnu/libnccl.so.2 /usr/lib/libnccl.so
# IMPORTANT:
# Add "ENV http_proxy=http://ip:port" if your download is slow, and don't forget to unset it at runtime.
# exmaple: unset http_proxy && unset https_proxy && python fluid_benchmark.py ...
RUN pip install -U pip
RUN pip install -U kubernetes paddlepaddle
RUN sh -c 'echo "import paddle.v2 as paddle\npaddle.dataset.cifar.train10()\npaddle.dataset.flowers.fetch()" | python'
RUN sh -c 'echo "import paddle.v2 as paddle\npaddle.dataset.mnist.train()\npaddle.dataset.mnist.test()\npaddle.dataset.imdb.fetch()" | python'
RUN sh -c 'echo "import paddle.v2 as paddle\npaddle.dataset.imikolov.fetch()" | python'
RUN pip uninstall -y paddlepaddle && mkdir /workspace
ADD https://raw.githubusercontent.com/PaddlePaddle/cloud/develop/docker/paddle_k8s /usr/bin
ADD https://raw.githubusercontent.com/PaddlePaddle/cloud/develop/docker/k8s_tools.py /root
RUN chmod +x /usr/bin/paddle_k8s
ADD *.whl /
RUN pip install /*.whl && rm -f /*.whl
ENV LD_LIBRARY_PATH=/usr/local/lib
ADD fluid_benchmark.py recordio_converter.py args.py recordio_converter.py run.sh run_fluid_benchmark.sh /workspace/
ADD models/ /workspace/models/
benchmark/fluid/README.md 0 → 100644
浏览文件 @ a38a8db9
# Fluid Benchmark
This directory contains several models configurations and tools that used to run
Fluid benchmarks for local and distributed training.
## Run the Benchmark
To start, run the following command to get the full help message:
```bash
python fluid_benchmark.py --help
```
Currently supported `--model` argument include:
* mnist
* resnet
* you can chose to use different dataset using `--data_set cifar10` or
`--data_set flowers`.
* vgg
* stacked_dynamic_lstm
* machine_translation
* Run the following command to start a benchmark job locally:
```bash
python fluid_benchmark.py --model mnist --device GPU
```
You can choose to use GPU/CPU training. With GPU training, you can specify
`--gpus <gpu_num>` to run multi GPU training.
You can set async mode parameter server. With async mode, you can specify
`--async_mode` to train model asynchronous.
* Run distributed training with parameter servers:
* see [run_fluid_benchmark.sh](https://github.com/PaddlePaddle/Paddle/blob/develop/benchmark/fluid/run_fluid_benchmark.sh) as an example.
* start parameter servers:
```bash
PADDLE_TRAINING_ROLE=PSERVER PADDLE_PSERVER_PORT=7164 PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=1 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=0 python fluid_benchmark.py --model mnist --device GPU --update_method pserver
sleep 15
```
* start trainers:
```bash
PADDLE_TRAINING_ROLE=TRAINER PADDLE_PSERVER_PORT=7164 PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=1 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=0 python fluid_benchmark.py --model mnist --device GPU --update_method pserver
```
* Run distributed training using NCCL2
```bash
PADDLE_PSERVER_PORT=7164 PADDLE_TRAINER_IPS=192.168.0.2,192.168.0.3 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=0 python fluid_benchmark.py --model mnist --device GPU --update_method nccl2
```
## Prepare the RecordIO file to Achieve Better Performance
Run the following command will generate RecordIO files like "mnist.recordio" under the path
and batch_size you choose, you can use batch_size=1 so that later reader can change the batch_size
at any time using `fluid.batch`.
```bash
python -c 'from recordio_converter import *; prepare_mnist("data", 1)'
```
## Run Distributed Benchmark on Kubernetes Cluster
You may need to build a Docker image before submitting a cluster job onto Kubernetes, or you will
have to start all those processes mannually on each node, which is not recommended.
To build the Docker image, you need to choose a paddle "whl" package to run with, you may either
download it from
http://www.paddlepaddle.org/docs/develop/documentation/zh/build_and_install/pip_install_en.html or
build it by your own. Once you've got the "whl" package, put it under the current directory and run:
```bash
docker build -t [your docker image name]:[your docker image tag] .
```
Then push the image to a Docker registry that your Kubernetes cluster can reach.
We provide a script `kube_gen_job.py` to generate Kubernetes yaml files to submit
distributed benchmark jobs to your cluster. To generate a job yaml, just run:
```bash
python kube_gen_job.py --jobname myjob --pscpu 4 --cpu 8 --gpu 8 --psmemory 20 --memory 40 --pservers 4 --trainers 4 --entry "python fluid_benchmark.py --model mnist --gpus 8 --device GPU --update_method pserver " --disttype pserver
```
Then the yaml files are generated under directory `myjob`, you can run:
```bash
kubectl create -f myjob/
```
The job shall start.
## Notes for Run Fluid Distributed with NCCL2 and RDMA
Before running NCCL2 distributed jobs, please check that whether your node has multiple network
interfaces, try to add the environment variable `export NCCL_SOCKET_IFNAME=eth0` to use your actual
network device.
To run high-performance distributed training, you must prepare your hardware environment to be
able to run RDMA enabled network communication, please check out [this](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/fluid/howto/cluster/nccl2_rdma_training.md)
note for details.
benchmark/fluid/args.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import argparse
__all__ = ['parse_args', ]
BENCHMARK_MODELS = [
"machine_translation", "resnet", "vgg", "mnist", "stacked_dynamic_lstm"
]
def parse_args():
parser = argparse.ArgumentParser('Fluid model benchmarks.')
parser.add_argument(
'--model',
type=str,
choices=BENCHMARK_MODELS,
default='resnet',
help='The model to run benchmark with.')
parser.add_argument(
'--batch_size', type=int, default=32, help='The minibatch size.')
# args related to learning rate
parser.add_argument(
'--learning_rate', type=float, default=0.001, help='The learning rate.')
# TODO(wuyi): add "--use_fake_data" option back.
parser.add_argument(
'--skip_batch_num',
type=int,
default=5,
help='The first num of minibatch num to skip, for better performance test'
)
parser.add_argument(
'--iterations', type=int, default=80, help='The number of minibatches.')
parser.add_argument(
'--pass_num', type=int, default=100, help='The number of passes.')
parser.add_argument(
'--data_format',
type=str,
default='NCHW',
choices=['NCHW', 'NHWC'],
help='The data data_format, now only support NCHW.')
parser.add_argument(
'--device',
type=str,
default='GPU',
choices=['CPU', 'GPU'],
help='The device type.')
parser.add_argument(
'--gpus',
type=int,
default=1,
help='If gpus > 1, will use ParallelExecutor to run, else use Executor.')
# this option is available only for vgg and resnet.
parser.add_argument(
'--cpus',
type=int,
default=1,
help='If cpus > 1, will use ParallelDo to run, else use Executor.')
parser.add_argument(
'--data_set',
type=str,
default='flowers',
choices=['cifar10', 'flowers'],
help='Optional dataset for benchmark.')
parser.add_argument(
'--infer_only', action='store_true', help='If set, run forward only.')
parser.add_argument(
'--use_cprof', action='store_true', help='If set, use cProfile.')
parser.add_argument(
'--use_nvprof',
action='store_true',
help='If set, use nvprof for CUDA.')
parser.add_argument(
'--no_test',
action='store_true',
help='If set, do not test the testset during training.')
parser.add_argument(
'--memory_optimize',
action='store_true',
help='If set, optimize runtime memory before start.')
parser.add_argument(
'--use_fake_data',
action='store_true',
help='If set ommit the actual read data operators.')
parser.add_argument(
'--profile', action='store_true', help='If set, profile a few steps.')
parser.add_argument(
'--update_method',
type=str,
default='local',
choices=['local', 'pserver', 'nccl2'],
help='Choose parameter update method, can be local, pserver, nccl2.')
parser.add_argument(
'--no_split_var',
action='store_true',
default=False,
help='Whether split variables into blocks when update_method is pserver')
parser.add_argument(
'--async_mode',
action='store_true',
default=False,
help='Whether start pserver in async mode to support ASGD')
parser.add_argument(
'--use_reader_op',
action='store_true',
help='Whether to use reader op, and must specify the data path if set this to true.'
)
parser.add_argument(
'--data_path',
type=str,
default="",
help='Directory that contains all the training recordio files.')
parser.add_argument(
'--use_inference_transpiler',
action='store_true',
help='If set, use inference transpiler to optimize the program.')
parser.add_argument(
'--no_random',
action='store_true',
help='If set, keep the random seed and do not shuffle the data.')
args = parser.parse_args()
return args
benchmark/fluid/fluid_benchmark.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import argparse
import cProfile
import time
import os
import numpy as np
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.profiler as profiler
import paddle.fluid.transpiler.distribute_transpiler as distribute_transpiler
from args import *
def append_nccl2_prepare(trainer_id):
if trainer_id >= 0:
# append gen_nccl_id at the end of startup program
trainer_id = int(os.getenv("PADDLE_TRAINER_ID"))
port = os.getenv("PADDLE_PSERVER_PORT")
worker_ips = os.getenv("PADDLE_TRAINER_IPS")
worker_endpoints = []
for ip in worker_ips.split(","):
worker_endpoints.append(':'.join([ip, port]))
num_trainers = len(worker_endpoints)
current_endpoint = os.getenv("PADDLE_CURRENT_IP") + ":" + port
worker_endpoints.remove(current_endpoint)
nccl_id_var = fluid.default_startup_program().global_block().create_var(
name="NCCLID",
persistable=True,
type=fluid.core.VarDesc.VarType.RAW)
fluid.default_startup_program().global_block().append_op(
type="gen_nccl_id",
inputs={},
outputs={"NCCLID": nccl_id_var},
attrs={
"endpoint": current_endpoint,
"endpoint_list": worker_endpoints,
"trainer_id": trainer_id
})
return nccl_id_var, num_trainers, trainer_id
else:
raise Exception("must set positive PADDLE_TRAINER_ID env variables for "
"nccl-based dist train.")
def dist_transpile(trainer_id, args):
if trainer_id < 0:
return None, None
# the port of all pservers, needed by both trainer and pserver
port = os.getenv("PADDLE_PSERVER_PORT", "6174")
# comma separated ips of all pservers, needed by trainer and
# pserver
pserver_ips = os.getenv("PADDLE_PSERVER_IPS", "")
eplist = []
for ip in pserver_ips.split(","):
eplist.append(':'.join([ip, port]))
pserver_endpoints = ",".join(eplist)
# total number of workers/trainers in the job, needed by
# trainer and pserver
trainers = int(os.getenv("PADDLE_TRAINERS"))
# the IP of the local machine, needed by pserver only
current_endpoint = os.getenv("PADDLE_CURRENT_IP", "") + ":" + port
# the role, should be either PSERVER or TRAINER
training_role = os.getenv("PADDLE_TRAINING_ROLE")
t = distribute_transpiler.DistributeTranspiler()
t.transpile(
trainer_id,
pservers=pserver_endpoints,
trainers=trainers,
sync_mode=not args.async_mode)
if training_role == "PSERVER":
pserver_program = t.get_pserver_program(current_endpoint)
pserver_startup_program = t.get_startup_program(current_endpoint,
pserver_program)
return pserver_program, pserver_startup_program
elif training_role == "TRAINER":
train_program = t.get_trainer_program()
return train_program, fluid.default_startup_program()
else:
raise ValueError(
'PADDLE_TRAINING_ROLE environment variable must be either TRAINER or PSERVER'
)
def test(exe, inference_program, test_reader, feeder, batch_acc):
accuracy_evaluator = fluid.metrics.Accuracy()
for batch_id, data in enumerate(test_reader()):
acc = exe.run(inference_program,
feed=feeder.feed(data),
fetch_list=[batch_acc])
accuracy_evaluator.update(value=np.array(acc), weight=len(data))
return accuracy_evaluator.eval()
# TODO(wuyi): replace train, train_parallel, test functions with new trainer
# API once it is ready.
def train(avg_loss, infer_prog, optimizer, train_reader, test_reader, batch_acc,
args, train_prog, startup_prog):
if os.getenv("PADDLE_TRAINING_ROLE") == "PSERVER":
place = core.CPUPlace()
exe = fluid.Executor(place)
exe.run(startup_prog)
exe.run(train_prog)
return
if args.use_fake_data:
raise Exception(
"fake data is not supported in single GPU test for now.")
place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(startup_prog)
# Use inference_transpiler to speedup
if not args.use_reader_op:
feed_var_list = [
var for var in train_prog.global_block().vars.itervalues()
if var.is_data
]
feeder = fluid.DataFeeder(feed_var_list, place)
iters, num_samples, start_time = 0, 0, time.time()
for pass_id in range(args.pass_num):
train_losses = []
if not args.use_reader_op:
reader_generator = train_reader()
batch_id = 0
data = None
while True:
if not args.use_reader_op:
data = next(reader_generator, None)
if data == None:
break
if iters == args.iterations:
break
if iters == args.skip_batch_num:
start_time = time.time()
num_samples = 0
if args.use_reader_op:
try:
loss = exe.run(train_prog, fetch_list=[avg_loss])
except fluid.core.EnforceNotMet as ex:
break
else:
loss = exe.run(train_prog,
feed=feeder.feed(data),
fetch_list=[avg_loss])
iters += 1
batch_id += 1
# FIXME(wuyi): For use_reader_op, if the current
# pass is not the last, the last batch of this pass
# is also equal to args.batch_size.
if args.use_reader_op:
num_samples += args.batch_size * args.gpus
else:
num_samples += len(data)
train_losses.append(loss)
print("Pass: %d, Iter: %d, Loss: %f\n" %
(pass_id, iters, np.mean(train_losses)))
print_train_time(start_time, time.time(), num_samples)
print("Pass: %d, Loss: %f" % (pass_id, np.mean(train_losses))),
# evaluation
if not args.no_test and batch_acc and not args.use_reader_op:
if args.use_inference_transpiler:
t = fluid.InferenceTranspiler()
t.transpile(infer_prog, place)
pass_test_acc = test(exe, infer_prog, test_reader, feeder,
batch_acc)
print(", Test Accuracy: %f" % pass_test_acc)
print("\n")
# TODO(wuyi): add warmup passes to get better perf data.
exit(0)
# TODO(wuyi): replace train, train_parallel, test functions with new trainer
# API once it is ready.
def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
batch_acc, args, train_prog, startup_prog, nccl_id_var,
num_trainers, trainer_id):
place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
if not args.use_reader_op:
feed_var_list = [
var for var in train_prog.global_block().vars.itervalues()
if var.is_data
]
feeder = fluid.DataFeeder(feed_var_list, place)
# generate fake:
if args.use_fake_data:
for var in feed_var_list:
v = startup_prog.global_block()._clone_variable(var)
var.persistable = True
v.persistable = True
real_shape = list(var.shape)
real_shape[0] = args.batch_size / args.gpus
startup_prog.global_block().append_op(
outputs={"Out": v},
type="fill_constant",
attrs={"shape": real_shape,
"value": 1.0,
"dtype": var.dtype})
if nccl_id_var and trainer_id == 0:
#FIXME(wuyi): wait other trainer to start listening
time.sleep(30)
startup_exe = fluid.Executor(place)
startup_exe.run(startup_prog)
strategy = fluid.ExecutionStrategy()
strategy.num_threads = 1
strategy.allow_op_delay = False
exe = fluid.ParallelExecutor(
True,
avg_loss.name,
exec_strategy=strategy,
num_trainers=num_trainers,
trainer_id=trainer_id)
for pass_id in range(args.pass_num):
num_samples = 0
iters = 0
start_time = time.time()
if not args.use_reader_op:
reader_generator = train_reader()
batch_id = 0
data = None
while True:
if not args.use_reader_op:
data = next(reader_generator, None)
if data == None:
break
if iters == args.iterations:
break
if args.profile and pass_id == 0 and batch_id == 5:
profiler.start_profiler("All")
elif args.profile and pass_id == 0 and batch_id == 10:
profiler.stop_profiler("total", "/tmp/profile_%d" % trainer_id)
if iters == args.skip_batch_num:
start_time = time.time()
num_samples = 0
if args.use_fake_data or args.use_reader_op:
try:
loss, = exe.run([avg_loss.name])
except fluid.core.EnforceNotMet as ex:
break
else:
loss, = exe.run([avg_loss.name], feed=feeder.feed(data))
if args.use_reader_op:
num_samples += args.batch_size * args.gpus
else:
num_samples += len(data)
iters += 1
if batch_id % 1 == 0:
print("Pass %d, batch %d, loss %s" %
(pass_id, batch_id, np.array(loss)))
batch_id += 1
print_train_time(start_time, time.time(), num_samples)
if not args.no_test and batch_acc and not args.use_reader_op:
# we have not implement record io for test
# skip test when use args.use_reader_op
test_acc = test(startup_exe, infer_prog, test_reader, feeder,
batch_acc)
print("Pass: %d, Test Accuracy: %f\n" % (pass_id, test_acc))
def print_arguments(args):
vars(args)['use_nvprof'] = (vars(args)['use_nvprof'] and
vars(args)['device'] == 'GPU')
print('----------- Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
def print_train_time(start_time, end_time, num_samples):
train_elapsed = end_time - start_time
examples_per_sec = num_samples / train_elapsed
print('\nTotal examples: %d, total time: %.5f, %.5f examples/sed\n' %
(num_samples, train_elapsed, examples_per_sec))
def print_paddle_envs():
print('----------- Configuration envs -----------')
for k in os.environ:
if "PADDLE_" in k:
print "ENV %s:%s" % (k, os.environ[k])
print('------------------------------------------------')
def main():
args = parse_args()
print_arguments(args)
print_paddle_envs()
if args.no_random:
fluid.default_startup_program().random_seed = 1
# the unique trainer id, starting from 0, needed by trainer
# only
nccl_id_var, num_trainers, trainer_id = (
None, 1, int(os.getenv("PADDLE_TRAINER_ID", "0")))
if args.use_cprof:
pr = cProfile.Profile()
pr.enable()
model_def = __import__("models.%s" % args.model, fromlist=["models"])
train_args = list(model_def.get_model(args))
train_args.append(args)
# Run optimizer.minimize(avg_loss)
train_args[2].minimize(train_args[0])
if args.memory_optimize:
fluid.memory_optimize(fluid.default_main_program())
if args.update_method == "pserver":
train_prog, startup_prog = dist_transpile(trainer_id, args)
if not train_prog:
raise Exception(
"Must configure correct environments to run dist train.")
train_args.extend([train_prog, startup_prog])
if args.gpus > 1 and os.getenv("PADDLE_TRAINING_ROLE") == "TRAINER":
train_args.extend([nccl_id_var, num_trainers, trainer_id])
train_parallel(*train_args)
train(*train_args)
exit(0)
# for other update methods, use default programs
train_args.append(fluid.default_main_program())
train_args.append(fluid.default_startup_program())
if args.update_method == "nccl2":
nccl_id_var, num_trainers, trainer_id = append_nccl2_prepare(trainer_id)
if args.gpus == 1:
# NOTE: parallel executor use profiler interanlly
if args.use_nvprof and args.device == 'GPU':
with profiler.cuda_profiler("cuda_profiler.txt", 'csv') as nvprof:
train(*train_args)
else:
train(*train_args)
else:
if args.device == "CPU":
raise Exception("Only support GPU perf with parallel exe")
train_args.extend([nccl_id_var, num_trainers, trainer_id])
train_parallel(*train_args)
if __name__ == "__main__":
main()
benchmark/fluid/kube_gen_job.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import yaml
import copy
import argparse
import random
import os
import copy
from kube_templates import pserver, trainer, envs
def parse_args():
parser = argparse.ArgumentParser(description='Generate dist job yamls.')
parser.add_argument(
'--jobname', default="paddlejob", help='unique job name')
parser.add_argument(
'--cpu', default=1, type=int, help='CPU cores per trainer node')
parser.add_argument(
'--pscpu', default=1, type=int, help='CPU cores per pserver node')
parser.add_argument(
'--gpu', default=0, type=int, help='num of GPUs per node')
parser.add_argument(
'--image',
default="bootstrapper:5000/fluid_benchmark:gpu",
help='num of GPUs per node')
parser.add_argument(
'--pservers', default=1, type=int, help='num of pservers')
parser.add_argument(
'--trainers', default=1, type=int, help='num of trainers')
parser.add_argument('--memory', default=1, type=int, help='trainer memory')
parser.add_argument(
'--psmemory', default=1, type=int, help='pserver memory')
parser.add_argument(
'--port', default=30236, type=int, help='num of trainers')
parser.add_argument(
'--entry', default="python train.py", help='command to run')
parser.add_argument(
'--fluid', default=1, type=int, help='whether is fluid job')
parser.add_argument(
'--rdma', action='store_true', help='whether mount rdma libs')
parser.add_argument(
'--disttype',
default="pserver",
type=str,
choices=['pserver', 'nccl2', 'local'],
help='pserver or nccl2 or local')
args = parser.parse_args()
return args
def gen_job():
ps = pserver
tn = trainer
args = parse_args()
ps_container = ps["spec"]["template"]["spec"]["containers"][0]
tn_container = tn["spec"]["template"]["spec"]["containers"][0]
if args.fluid == 1:
ps_container["command"] = \
["paddle_k8s", "start_fluid"]
tn_container["command"] = \
["paddle_k8s", "start_fluid"]
ps["metadata"]["name"] = args.jobname + "-pserver"
ps["spec"]["template"]["metadata"]["labels"][
"paddle-job-pserver"] = args.jobname
tn["metadata"]["name"] = args.jobname + "-trainer"
tn["spec"]["template"]["metadata"]["labels"]["paddle-job"] = args.jobname
ps_container["image"] = args.image
tn_container["image"] = args.image
ps_container["resources"]["requests"]["cpu"] = str(args.pscpu)
ps_container["resources"]["requests"]["memory"] = str(args.psmemory) + "Gi"
ps_container["resources"]["limits"]["cpu"] = str(args.pscpu)
ps_container["resources"]["limits"]["memory"] = str(args.psmemory) + "Gi"
tn_container["resources"]["requests"]["cpu"] = str(args.cpu)
tn_container["resources"]["requests"]["memory"] = str(args.memory) + "Gi"
tn_container["resources"]["limits"]["cpu"] = str(args.cpu)
tn_container["resources"]["limits"]["memory"] = str(args.memory) + "Gi"
if args.gpu > 0:
tn_container["resources"]["requests"][
"alpha.kubernetes.io/nvidia-gpu"] = str(args.gpu)
tn_container["resources"]["limits"][
"alpha.kubernetes.io/nvidia-gpu"] = str(args.gpu)
ps["spec"]["replicas"] = int(args.pservers)
tn["spec"]["parallelism"] = int(args.trainers)
tn["spec"]["completions"] = int(args.trainers)
ps_container["ports"][0]["name"] = "jobport-" + str(args.port)
ps_container["ports"][0]["containerPort"] = args.port
spreadport = random.randint(40000, 60000)
tn_container["ports"][0]["name"] = "spr-" + str(spreadport)
tn_container["ports"][0]["containerPort"] = spreadport
envs.append({"name": "PADDLE_JOB_NAME", "value": args.jobname})
envs.append({"name": "PADDLE_TRAINERS", "value": str(args.trainers)})
envs.append({"name": "PADDLE_PSERVERS", "value": str(args.pservers)})
envs.append({"name": "ENTRY", "value": args.entry})
envs.append({"name": "PADDLE_PSERVER_PORT", "value": str(args.port)})
# NOTE: these directories below are cluster specific, please modify
# this settings before you run on your own cluster.
envs.append({
"name": "LD_LIBRARY_PATH",
"value":
"/usr/local/lib:/usr/local/nvidia/lib64:/usr/local/rdma/lib64:/usr/lib64/mlnx_ofed/valgrind"
})
volumes = [{
"name": "nvidia-driver",
"hostPath": {
"path": "/usr/local/nvidia/lib64"
}
}]
volumeMounts = [{
"mountPath": "/usr/local/nvidia/lib64",
"name": "nvidia-driver"
}]
if args.rdma:
volumes.extend([{
"name": "ibetc",
"hostPath": {
"path": "/etc/libibverbs.d"
}
}, {
"name": "iblibs",
"hostPath": {
"path": "/usr/local/rdma"
}
}, {
"name": "valgrind",
"hostPath": {
"path": "/usr/lib64/mlnx_ofed/valgrind"
}
}])
volumeMounts.extend([{
"mountPath": "/etc/libibverbs.d",
"name": "ibetc"
}, {
"mountPath": "/usr/local/rdma",
"name": "iblibs"
}, {
"mountPath": "/usr/lib64/mlnx_ofed/valgrind",
"name": "valgrind"
}])
# append shm for NCCL2
volumes.append({"name": "dshm", "emptyDir": {"medium": "Memory"}})
volumeMounts.append({"mountPath": "/dev/shm", "name": "dshm"})
tn["spec"]["template"]["spec"]["volumes"] = volumes
tn_container["volumeMounts"] = volumeMounts
ps_container["env"] = copy.deepcopy(envs)
ps_container["env"].append({
"name": "PADDLE_TRAINING_ROLE",
"value": "PSERVER"
})
tn_container["env"] = envs
if args.disttype == "pserver":
tn_container["env"].append({
"name": "PADDLE_TRAINING_ROLE",
"value": "TRAINER"
})
elif args.disttype == "nccl2" or args.disttype == "local":
# NCCL2 have no training role, set to plain WORKER
tn_container["env"].append({
"name": "PADDLE_TRAINING_ROLE",
"value": "WORKER"
})
os.mkdir(args.jobname)
if args.disttype == "pserver":
with open("%s/pserver.yaml" % args.jobname, "w") as fn:
yaml.dump(ps, fn)
with open("%s/trainer.yaml" % args.jobname, "w") as fn:
yaml.dump(tn, fn)
if __name__ == "__main__":
gen_job()
benchmark/fluid/kube_templates/__init__.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from pserver import pserver
from trainer import trainer
__all__ = ["pserver", "trainer", "envs"]
envs = [
# envs that don't need to change
{
"name": "GLOG_v",
"value": "0"
},
{
"name": "GLOG_logtostderr",
"value": "1"
},
{
"name": "TOPOLOGY",
"value": ""
},
{
"name": "TRAINER_PACKAGE",
"value": "/workspace"
},
{
"name": "PADDLE_INIT_NICS",
"value": "eth2"
},
{
"name": "NAMESPACE",
"valueFrom": {
"fieldRef": {
"fieldPath": "metadata.namespace"
}
}
},
{
"name": "POD_IP",
"valueFrom": {
"fieldRef": {
"fieldPath": "status.podIP"
}
}
},
{
"name": "PADDLE_CURRENT_IP",
"valueFrom": {
"fieldRef": {
"fieldPath": "status.podIP"
}
}
}
]
benchmark/fluid/kube_templates/pserver.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
pserver = {
"apiVersion": "extensions/v1beta1",
"kind": "ReplicaSet",
"metadata": {
"name": "jobname-pserver"
},
"spec": {
"replicas": 1,
"template": {
"metadata": {
"labels": {
"paddle-job-pserver": "jobname"
}
},
"spec": {
"hostNetwork": True,
"imagePullSecrets": [{
"name": "job-registry-secret"
}],
"containers": [{
"name": "pserver",
"image": "",
"imagePullPolicy": "Always",
"ports": [{
"name": "jobport-1",
"containerPort": 1
}],
"env": [],
"command": ["paddle_k8s", "start_pserver"],
"resources": {
"requests": {
"memory": "10Gi",
"cpu": "4"
},
"limits": {
"memory": "10Gi",
"cpu": "4"
}
}
}]
}
}
}
}
benchmark/fluid/kube_templates/trainer.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
trainer = {
"apiVersion": "batch/v1",
"kind": "Job",
"metadata": {
"name": "jobname-pserver"
},
"spec": {
"parallelism": 4,
"completions": 4,
"template": {
"metadata": {
"labels": {
"paddle-job": "jobname"
}
},
"spec": {
"hostNetwork": True,
"imagePullSecrets": [{
"name": "job-registry-secret"
}],
"restartPolicy": "Never",
"containers": [{
"name": "trainer",
"image": "",
"imagePullPolicy": "Always",
# to let container set rlimit
"securityContext": {
"privileged": True
# TODO(wuyi): use below specific cap instead of privileged,
# using privileged will cause all GPU device are visible
# in the container.
# "capabilities": {
# "add": ["SYS_RESOURCE"]
# }
},
"ports": [{
"name": "jobport-1",
"containerPort": 1
}],
"env": [],
"command": ["paddle_k8s", "start_trainer", "v2"],
"resources": {
"requests": {
"memory": "10Gi",
"cpu": "4",
},
"limits": {
"memory": "10Gi",
"cpu": "4",
}
}
}]
}
}
}
}
benchmark/fluid/machine_translation.py 已删除 100644 → 0
浏览文件 @ 10ec329b
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""seq2seq model for fluid."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import argparse
import time
import distutils.util
import paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.framework as framework
from paddle.fluid.executor import Executor
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
"--embedding_dim",
type=int,
default=512,
help="The dimension of embedding table. (default: %(default)d)")
parser.add_argument(
"--encoder_size",
type=int,
default=512,
help="The size of encoder bi-rnn unit. (default: %(default)d)")
parser.add_argument(
"--decoder_size",
type=int,
default=512,
help="The size of decoder rnn unit. (default: %(default)d)")
parser.add_argument(
"--batch_size",
type=int,
default=16,
help="The sequence number of a mini-batch data. (default: %(default)d)")
parser.add_argument(
'--skip_batch_num',
type=int,
default=5,
help='The first num of minibatch num to skip, for better performance test')
parser.add_argument(
'--iterations', type=int, default=80, help='The number of minibatches.')
parser.add_argument(
"--dict_size",
type=int,
default=30000,
help="The dictionary capacity. Dictionaries of source sequence and "
"target dictionary have same capacity. (default: %(default)d)")
parser.add_argument(
"--pass_num",
type=int,
default=2,
help="The pass number to train. (default: %(default)d)")
parser.add_argument(
"--learning_rate",
type=float,
default=0.0002,
help="Learning rate used to train the model. (default: %(default)f)")
parser.add_argument(
"--infer_only", action='store_true', help="If set, run forward only.")
parser.add_argument(
"--beam_size",
type=int,
default=3,
help="The width for beam searching. (default: %(default)d)")
parser.add_argument(
'--device',
type=str,
default='GPU',
choices=['CPU', 'GPU'],
help="The device type.")
parser.add_argument(
"--max_length",
type=int,
default=250,
help="The maximum length of sequence when doing generation. "
"(default: %(default)d)")
parser.add_argument(
'--with_test',
action='store_true',
help='If set, test the testset during training.')
def lstm_step(x_t, hidden_t_prev, cell_t_prev, size):
def linear(inputs):
return fluid.layers.fc(input=inputs, size=size, bias_attr=True)
forget_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
input_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
output_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
cell_tilde = fluid.layers.tanh(x=linear([hidden_t_prev, x_t]))
cell_t = fluid.layers.sums(input=[
fluid.layers.elementwise_mul(
x=forget_gate, y=cell_t_prev), fluid.layers.elementwise_mul(
x=input_gate, y=cell_tilde)
])
hidden_t = fluid.layers.elementwise_mul(
x=output_gate, y=fluid.layers.tanh(x=cell_t))
return hidden_t, cell_t
def seq_to_seq_net(embedding_dim, encoder_size, decoder_size, source_dict_dim,
target_dict_dim, is_generating, beam_size, max_length):
"""Construct a seq2seq network."""
def bi_lstm_encoder(input_seq, gate_size):
# Linear transformation part for input gate, output gate, forget gate
# and cell activation vectors need be done outside of dynamic_lstm.
# So the output size is 4 times of gate_size.
input_forward_proj = fluid.layers.fc(input=input_seq,
size=gate_size * 4,
act=None,
bias_attr=False)
forward, _ = fluid.layers.dynamic_lstm(
input=input_forward_proj, size=gate_size * 4, use_peepholes=False)
input_reversed_proj = fluid.layers.fc(input=input_seq,
size=gate_size * 4,
act=None,
bias_attr=False)
reversed, _ = fluid.layers.dynamic_lstm(
input=input_reversed_proj,
size=gate_size * 4,
is_reverse=True,
use_peepholes=False)
return forward, reversed
src_word_idx = fluid.layers.data(
name='source_sequence', shape=[1], dtype='int64', lod_level=1)
src_embedding = fluid.layers.embedding(
input=src_word_idx,
size=[source_dict_dim, embedding_dim],
dtype='float32')
src_forward, src_reversed = bi_lstm_encoder(
input_seq=src_embedding, gate_size=encoder_size)
encoded_vector = fluid.layers.concat(
input=[src_forward, src_reversed], axis=1)
encoded_proj = fluid.layers.fc(input=encoded_vector,
size=decoder_size,
bias_attr=False)
backward_first = fluid.layers.sequence_pool(
input=src_reversed, pool_type='first')
decoder_boot = fluid.layers.fc(input=backward_first,
size=decoder_size,
bias_attr=False,
act='tanh')
def lstm_decoder_with_attention(target_embedding, encoder_vec, encoder_proj,
decoder_boot, decoder_size):
def simple_attention(encoder_vec, encoder_proj, decoder_state):
decoder_state_proj = fluid.layers.fc(input=decoder_state,
size=decoder_size,
bias_attr=False)
decoder_state_expand = fluid.layers.sequence_expand(
x=decoder_state_proj, y=encoder_proj)
concated = fluid.layers.concat(
input=[encoder_proj, decoder_state_expand], axis=1)
attention_weights = fluid.layers.fc(input=concated,
size=1,
act='tanh',
bias_attr=False)
attention_weights = fluid.layers.sequence_softmax(
input=attention_weights)
weigths_reshape = fluid.layers.reshape(
x=attention_weights, shape=[-1])
scaled = fluid.layers.elementwise_mul(
x=encoder_vec, y=weigths_reshape, axis=0)
context = fluid.layers.sequence_pool(input=scaled, pool_type='sum')
return context
rnn = fluid.layers.DynamicRNN()
cell_init = fluid.layers.fill_constant_batch_size_like(
input=decoder_boot,
value=0.0,
shape=[-1, decoder_size],
dtype='float32')
cell_init.stop_gradient = False
with rnn.block():
current_word = rnn.step_input(target_embedding)
encoder_vec = rnn.static_input(encoder_vec)
encoder_proj = rnn.static_input(encoder_proj)
hidden_mem = rnn.memory(init=decoder_boot, need_reorder=True)
cell_mem = rnn.memory(init=cell_init)
context = simple_attention(encoder_vec, encoder_proj, hidden_mem)
decoder_inputs = fluid.layers.concat(
input=[context, current_word], axis=1)
h, c = lstm_step(decoder_inputs, hidden_mem, cell_mem, decoder_size)
rnn.update_memory(hidden_mem, h)
rnn.update_memory(cell_mem, c)
out = fluid.layers.fc(input=h,
size=target_dict_dim,
bias_attr=True,
act='softmax')
rnn.output(out)
return rnn()
if not is_generating:
trg_word_idx = fluid.layers.data(
name='target_sequence', shape=[1], dtype='int64', lod_level=1)
trg_embedding = fluid.layers.embedding(
input=trg_word_idx,
size=[target_dict_dim, embedding_dim],
dtype='float32')
prediction = lstm_decoder_with_attention(trg_embedding, encoded_vector,
encoded_proj, decoder_boot,
decoder_size)
label = fluid.layers.data(
name='label_sequence', shape=[1], dtype='int64', lod_level=1)
cost = fluid.layers.cross_entropy(input=prediction, label=label)
avg_cost = fluid.layers.mean(x=cost)
feeding_list = ["source_sequence", "target_sequence", "label_sequence"]
return avg_cost, feeding_list
def to_lodtensor(data, place):
seq_lens = [len(seq) for seq in data]
cur_len = 0
lod = [cur_len]
for l in seq_lens:
cur_len += l
lod.append(cur_len)
flattened_data = np.concatenate(data, axis=0).astype("int64")
flattened_data = flattened_data.reshape([len(flattened_data), 1])
lod_t = core.LoDTensor()
lod_t.set(flattened_data, place)
lod_t.set_lod([lod])
return lod_t, lod[-1]
def lodtensor_to_ndarray(lod_tensor):
dims = lod_tensor.get_dims()
ndarray = np.zeros(shape=dims).astype('float32')
for i in xrange(np.product(dims)):
ndarray.ravel()[i] = lod_tensor.get_float_element(i)
return ndarray
def train():
avg_cost, feeding_list = seq_to_seq_net(
args.embedding_dim,
args.encoder_size,
args.decoder_size,
args.dict_size,
args.dict_size,
False,
beam_size=args.beam_size,
max_length=args.max_length)
# clone from default main program
inference_program = fluid.default_main_program().clone()
optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
optimizer.minimize(avg_cost)
fluid.memory_optimize(fluid.default_main_program())
train_batch_generator = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.wmt14.train(args.dict_size), buf_size=1000),
batch_size=args.batch_size)
test_batch_generator = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.wmt14.test(args.dict_size), buf_size=1000),
batch_size=args.batch_size)
place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
exe = Executor(place)
exe.run(framework.default_startup_program())
def do_validation():
total_loss = 0.0
count = 0
for batch_id, data in enumerate(test_batch_generator()):
src_seq = to_lodtensor(map(lambda x: x[0], data), place)[0]
trg_seq = to_lodtensor(map(lambda x: x[1], data), place)[0]
lbl_seq = to_lodtensor(map(lambda x: x[2], data), place)[0]
fetch_outs = exe.run(inference_program,
feed={
feeding_list[0]: src_seq,
feeding_list[1]: trg_seq,
feeding_list[2]: lbl_seq
},
fetch_list=[avg_cost],
return_numpy=False)
total_loss += lodtensor_to_ndarray(fetch_outs[0])[0]
count += 1
return total_loss / count
iters, num_samples, start_time = 0, 0, time.time()
for pass_id in xrange(args.pass_num):
train_accs = []
train_losses = []
for batch_id, data in enumerate(train_batch_generator()):
if iters == args.skip_batch_num:
start_time = time.time()
num_samples = 0
if iters == args.iterations:
break
src_seq, word_num = to_lodtensor(map(lambda x: x[0], data), place)
num_samples += word_num
trg_seq, word_num = to_lodtensor(map(lambda x: x[1], data), place)
num_samples += word_num
lbl_seq, _ = to_lodtensor(map(lambda x: x[2], data), place)
fetch_outs = exe.run(framework.default_main_program(),
feed={
feeding_list[0]: src_seq,
feeding_list[1]: trg_seq,
feeding_list[2]: lbl_seq
},
fetch_list=[avg_cost])
iters += 1
loss = np.array(fetch_outs[0])
print(
"Pass = %d, Iter = %d, Loss = %f" % (pass_id, iters, loss)
) # The accuracy is the accumulation of batches, but not the current batch.
train_elapsed = time.time() - start_time
examples_per_sec = num_samples / train_elapsed
print('\nTotal examples: %d, total time: %.5f, %.5f examples/sed\n' %
(num_samples, train_elapsed, examples_per_sec))
# evaluation
if args.with_test:
test_loss = do_validation()
exit(0)
def infer():
pass
def print_arguments(args):
print('----------- seq2seq Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
if __name__ == '__main__':
args = parser.parse_args()
print_arguments(args)
if args.infer_only:
infer()
else:
train()
benchmark/fluid/mnist.py 已删除 100644 → 0
浏览文件 @ 10ec329b
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import argparse
import time
import paddle
import paddle.fluid as fluid
import paddle.fluid.profiler as profiler
SEED = 1
DTYPE = "float32"
# random seed must set before configuring the network.
# fluid.default_startup_program().random_seed = SEED
def parse_args():
parser = argparse.ArgumentParser("mnist model benchmark.")
parser.add_argument(
'--batch_size', type=int, default=128, help='The minibatch size.')
parser.add_argument(
'--skip_batch_num',
type=int,
default=5,
help='The first num of minibatch num to skip, for better performance test'
)
parser.add_argument(
'--iterations', type=int, default=35, help='The number of minibatches.')
parser.add_argument(
'--pass_num', type=int, default=5, help='The number of passes.')
parser.add_argument(
'--device',
type=str,
default='GPU',
choices=['CPU', 'GPU'],
help='The device type.')
parser.add_argument(
'--infer_only', action='store_true', help='If set, run forward only.')
parser.add_argument(
'--use_cprof', action='store_true', help='If set, use cProfile.')
parser.add_argument(
'--use_nvprof',
action='store_true',
help='If set, use nvprof for CUDA.')
parser.add_argument(
'--with_test',
action='store_true',
help='If set, test the testset during training.')
args = parser.parse_args()
return args
def cnn_model(data):
conv_pool_1 = fluid.nets.simple_img_conv_pool(
input=data,
filter_size=5,
num_filters=20,
pool_size=2,
pool_stride=2,
act="relu")
conv_pool_2 = fluid.nets.simple_img_conv_pool(
input=conv_pool_1,
filter_size=5,
num_filters=50,
pool_size=2,
pool_stride=2,
act="relu")
# TODO(dzhwinter) : refine the initializer and random seed settting
SIZE = 10
input_shape = conv_pool_2.shape
param_shape = [reduce(lambda a, b: a * b, input_shape[1:], 1)] + [SIZE]
scale = (2.0 / (param_shape[0]**2 * SIZE))**0.5
predict = fluid.layers.fc(
input=conv_pool_2,
size=SIZE,
act="softmax",
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.NormalInitializer(
loc=0.0, scale=scale)))
return predict
def eval_test(exe, batch_acc, batch_size_tensor, inference_program):
test_reader = paddle.batch(
paddle.dataset.mnist.test(), batch_size=args.batch_size)
test_pass_acc = fluid.average.WeightedAverage()
for batch_id, data in enumerate(test_reader()):
img_data = np.array(map(lambda x: x[0].reshape([1, 28, 28]),
data)).astype(DTYPE)
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([len(y_data), 1])
acc, weight = exe.run(inference_program,
feed={"pixel": img_data,
"label": y_data},
fetch_list=[batch_acc, batch_size_tensor])
test_pass_acc.add(value=acc, weight=weight)
pass_acc = test_pass_acc.eval()
return pass_acc
def run_benchmark(model, args):
if args.use_cprof:
pr = cProfile.Profile()
pr.enable()
start_time = time.time()
# Input data
images = fluid.layers.data(name='pixel', shape=[1, 28, 28], dtype=DTYPE)
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
# Train program
predict = model(images)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
# Evaluator
batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(
input=predict, label=label, total=batch_size_tensor)
# inference program
inference_program = fluid.default_main_program().clone()
# Optimization
opt = fluid.optimizer.AdamOptimizer(
learning_rate=0.001, beta1=0.9, beta2=0.999)
opt.minimize(avg_cost)
fluid.memory_optimize(fluid.default_main_program())
# Initialize executor
place = fluid.CPUPlace() if args.device == 'CPU' else fluid.CUDAPlace(0)
exe = fluid.Executor(place)
# Parameter initialization
exe.run(fluid.default_startup_program())
# Reader
train_reader = paddle.batch(
paddle.dataset.mnist.train(), batch_size=args.batch_size)
accuracy = fluid.metrics.Accuracy()
iters, num_samples, start_time = 0, 0, time.time()
for pass_id in range(args.pass_num):
accuracy.reset()
train_accs = []
train_losses = []
for batch_id, data in enumerate(train_reader()):
if iters == args.skip_batch_num:
start_time = time.time()
num_samples = 0
if iters == args.iterations:
break
img_data = np.array(
map(lambda x: x[0].reshape([1, 28, 28]), data)).astype(DTYPE)
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([len(y_data), 1])
outs = exe.run(
fluid.default_main_program(),
feed={"pixel": img_data,
"label": y_data},
fetch_list=[avg_cost, batch_acc, batch_size_tensor]
) # The accuracy is the accumulation of batches, but not the current batch.
accuracy.update(value=outs[1], weight=outs[2])
iters += 1
num_samples += len(y_data)
loss = np.array(outs[0])
acc = np.array(outs[1])
train_losses.append(loss)
train_accs.append(acc)
print("Pass: %d, Iter: %d, Loss: %f, Accuracy: %f" %
(pass_id, iters, loss, acc))
print("Pass: %d, Loss: %f, Train Accuray: %f\n" %
(pass_id, np.mean(train_losses), np.mean(train_accs)))
train_elapsed = time.time() - start_time
examples_per_sec = num_samples / train_elapsed
print('\nTotal examples: %d, total time: %.5f, %.5f examples/sed\n' %
(num_samples, train_elapsed, examples_per_sec))
# evaluation
if args.with_test:
test_avg_acc = eval_test(exe, batch_acc, batch_size_tensor,
inference_program)
exit(0)
def print_arguments(args):
vars(args)['use_nvprof'] = (vars(args)['use_nvprof'] and
vars(args)['device'] == 'GPU')
print('----------- mnist Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
if __name__ == '__main__':
args = parse_args()
print_arguments(args)
if args.use_nvprof and args.device == 'GPU':
with profiler.cuda_profiler("cuda_profiler.txt", 'csv') as nvprof:
run_benchmark(cnn_model, args)
else:
run_benchmark(cnn_model, args)
benchmark/fluid/models/__init__.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
__all__ = [
"machine_translation", "resnet", "vgg", "mnist", "stacked_dynamic_lstm"
]
benchmark/fluid/models/machine_translation.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""seq2seq model for fluid."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import argparse
import time
import distutils.util
import paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.framework as framework
from paddle.fluid.executor import Executor
def lstm_step(x_t, hidden_t_prev, cell_t_prev, size):
def linear(inputs):
return fluid.layers.fc(input=inputs, size=size, bias_attr=True)
forget_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
input_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
output_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t]))
cell_tilde = fluid.layers.tanh(x=linear([hidden_t_prev, x_t]))
cell_t = fluid.layers.sums(input=[
fluid.layers.elementwise_mul(
x=forget_gate, y=cell_t_prev), fluid.layers.elementwise_mul(
x=input_gate, y=cell_tilde)
])
hidden_t = fluid.layers.elementwise_mul(
x=output_gate, y=fluid.layers.tanh(x=cell_t))
return hidden_t, cell_t
def seq_to_seq_net(embedding_dim, encoder_size, decoder_size, source_dict_dim,
target_dict_dim, is_generating, beam_size, max_length):
"""Construct a seq2seq network."""
def bi_lstm_encoder(input_seq, gate_size):
# Linear transformation part for input gate, output gate, forget gate
# and cell activation vectors need be done outside of dynamic_lstm.
# So the output size is 4 times of gate_size.
input_forward_proj = fluid.layers.fc(input=input_seq,
size=gate_size * 4,
act=None,
bias_attr=False)
forward, _ = fluid.layers.dynamic_lstm(
input=input_forward_proj, size=gate_size * 4, use_peepholes=False)
input_reversed_proj = fluid.layers.fc(input=input_seq,
size=gate_size * 4,
act=None,
bias_attr=False)
reversed, _ = fluid.layers.dynamic_lstm(
input=input_reversed_proj,
size=gate_size * 4,
is_reverse=True,
use_peepholes=False)
return forward, reversed
src_word_idx = fluid.layers.data(
name='source_sequence', shape=[1], dtype='int64', lod_level=1)
src_embedding = fluid.layers.embedding(
input=src_word_idx,
size=[source_dict_dim, embedding_dim],
dtype='float32')
src_forward, src_reversed = bi_lstm_encoder(
input_seq=src_embedding, gate_size=encoder_size)
encoded_vector = fluid.layers.concat(
input=[src_forward, src_reversed], axis=1)
encoded_proj = fluid.layers.fc(input=encoded_vector,
size=decoder_size,
bias_attr=False)
backward_first = fluid.layers.sequence_pool(
input=src_reversed, pool_type='first')
decoder_boot = fluid.layers.fc(input=backward_first,
size=decoder_size,
bias_attr=False,
act='tanh')
def lstm_decoder_with_attention(target_embedding, encoder_vec, encoder_proj,
decoder_boot, decoder_size):
def simple_attention(encoder_vec, encoder_proj, decoder_state):
decoder_state_proj = fluid.layers.fc(input=decoder_state,
size=decoder_size,
bias_attr=False)
decoder_state_expand = fluid.layers.sequence_expand(
x=decoder_state_proj, y=encoder_proj)
concated = fluid.layers.concat(
input=[encoder_proj, decoder_state_expand], axis=1)
attention_weights = fluid.layers.fc(input=concated,
size=1,
act='tanh',
bias_attr=False)
attention_weights = fluid.layers.sequence_softmax(
input=attention_weights)
weigths_reshape = fluid.layers.reshape(
x=attention_weights, shape=[-1])
scaled = fluid.layers.elementwise_mul(
x=encoder_vec, y=weigths_reshape, axis=0)
context = fluid.layers.sequence_pool(input=scaled, pool_type='sum')
return context
rnn = fluid.layers.DynamicRNN()
cell_init = fluid.layers.fill_constant_batch_size_like(
input=decoder_boot,
value=0.0,
shape=[-1, decoder_size],
dtype='float32')
cell_init.stop_gradient = False
with rnn.block():
current_word = rnn.step_input(target_embedding)
encoder_vec = rnn.static_input(encoder_vec)
encoder_proj = rnn.static_input(encoder_proj)
hidden_mem = rnn.memory(init=decoder_boot, need_reorder=True)
cell_mem = rnn.memory(init=cell_init)
context = simple_attention(encoder_vec, encoder_proj, hidden_mem)
decoder_inputs = fluid.layers.concat(
input=[context, current_word], axis=1)
h, c = lstm_step(decoder_inputs, hidden_mem, cell_mem, decoder_size)
rnn.update_memory(hidden_mem, h)
rnn.update_memory(cell_mem, c)
out = fluid.layers.fc(input=h,
size=target_dict_dim,
bias_attr=True,
act='softmax')
rnn.output(out)
return rnn()
if not is_generating:
trg_word_idx = fluid.layers.data(
name='target_sequence', shape=[1], dtype='int64', lod_level=1)
trg_embedding = fluid.layers.embedding(
input=trg_word_idx,
size=[target_dict_dim, embedding_dim],
dtype='float32')
prediction = lstm_decoder_with_attention(trg_embedding, encoded_vector,
encoded_proj, decoder_boot,
decoder_size)
label = fluid.layers.data(
name='label_sequence', shape=[1], dtype='int64', lod_level=1)
cost = fluid.layers.cross_entropy(input=prediction, label=label)
avg_cost = fluid.layers.mean(x=cost)
feeding_list = ["source_sequence", "target_sequence", "label_sequence"]
return avg_cost, feeding_list
def lodtensor_to_ndarray(lod_tensor):
dims = lod_tensor.get_dims()
ndarray = np.zeros(shape=dims).astype('float32')
for i in xrange(np.product(dims)):
ndarray.ravel()[i] = lod_tensor.get_float_element(i)
return ndarray
def get_model(args):
if args.use_reader_op:
raise Exception("machine_translation do not support reader op for now.")
embedding_dim = 512
encoder_size = 512
decoder_size = 512
dict_size = 30000
beam_size = 3
max_length = 250
avg_cost, feeding_list = seq_to_seq_net(
embedding_dim,
encoder_size,
decoder_size,
dict_size,
dict_size,
False,
beam_size=beam_size,
max_length=max_length)
# clone from default main program
inference_program = fluid.default_main_program().clone()
optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
train_batch_generator = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.wmt14.train(dict_size), buf_size=1000),
batch_size=args.batch_size * args.gpus)
test_batch_generator = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.wmt14.test(dict_size), buf_size=1000),
batch_size=args.batch_size)
return avg_cost, inference_program, optimizer, train_batch_generator, \
test_batch_generator, None
benchmark/fluid/models/mnist.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import argparse
import time
import cProfile
import os
import paddle
import paddle.fluid as fluid
import paddle.fluid.profiler as profiler
SEED = 1
DTYPE = "float32"
# random seed must set before configuring the network.
# fluid.default_startup_program().random_seed = SEED
def cnn_model(data):
conv_pool_1 = fluid.nets.simple_img_conv_pool(
input=data,
filter_size=5,
num_filters=20,
pool_size=2,
pool_stride=2,
act="relu")
conv_pool_2 = fluid.nets.simple_img_conv_pool(
input=conv_pool_1,
filter_size=5,
num_filters=50,
pool_size=2,
pool_stride=2,
act="relu")
# TODO(dzhwinter) : refine the initializer and random seed settting
SIZE = 10
input_shape = conv_pool_2.shape
param_shape = [reduce(lambda a, b: a * b, input_shape[1:], 1)] + [SIZE]
scale = (2.0 / (param_shape[0]**2 * SIZE))**0.5
predict = fluid.layers.fc(
input=conv_pool_2,
size=SIZE,
act="softmax",
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.NormalInitializer(
loc=0.0, scale=scale)))
return predict
def get_model(args):
if args.use_reader_op:
filelist = [
os.path.join(args.data_path, f) for f in os.listdir(args.data_path)
]
data_file = fluid.layers.open_files(
filenames=filelist,
shapes=[[-1, 1, 28, 28], (-1, 1)],
lod_levels=[0, 0],
dtypes=["float32", "int64"],
thread_num=args.gpus,
pass_num=args.pass_num)
data_file = fluid.layers.double_buffer(
fluid.layers.batch(
data_file, batch_size=args.batch_size))
images, label = fluid.layers.read_file(data_file)
else:
images = fluid.layers.data(name='pixel', shape=[1, 28, 28], dtype=DTYPE)
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
if args.device == 'CPU' and args.cpus > 1:
places = fluid.layers.get_places(args.cpus)
pd = fluid.layers.ParallelDo(places)
with pd.do():
predict = cnn_model(pd.read_input(images))
label = pd.read_input(label)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
batch_acc = fluid.layers.accuracy(input=predict, label=label)
pd.write_output(avg_cost)
pd.write_output(batch_acc)
avg_cost, batch_acc = pd()
avg_cost = fluid.layers.mean(avg_cost)
batch_acc = fluid.layers.mean(batch_acc)
else:
# Train program
predict = cnn_model(images)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
# Evaluator
batch_acc = fluid.layers.accuracy(input=predict, label=label)
# inference program
inference_program = fluid.default_main_program().clone()
# Optimization
opt = fluid.optimizer.AdamOptimizer(
learning_rate=0.001, beta1=0.9, beta2=0.999)
# Reader
train_reader = paddle.batch(
paddle.dataset.mnist.train(), batch_size=args.batch_size * args.gpus)
test_reader = paddle.batch(
paddle.dataset.mnist.test(), batch_size=args.batch_size)
return avg_cost, inference_program, opt, train_reader, test_reader, batch_acc
benchmark/fluid/models/resnet.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import functools
import numpy as np
import time
import os
import cProfile, pstats, StringIO
import paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.profiler as profiler
from recordio_converter import imagenet_train, imagenet_test
def conv_bn_layer(input, ch_out, filter_size, stride, padding, act='relu'):
conv1 = fluid.layers.conv2d(
input=input,
filter_size=filter_size,
num_filters=ch_out,
stride=stride,
padding=padding,
act=None,
bias_attr=False)
return fluid.layers.batch_norm(input=conv1, act=act)
def shortcut(input, ch_out, stride):
ch_in = input.shape[1] # if args.data_format == 'NCHW' else input.shape[-1]
if ch_in != ch_out:
return conv_bn_layer(input, ch_out, 1, stride, 0, None)
else:
return input
def basicblock(input, ch_out, stride):
short = shortcut(input, ch_out, stride)
conv1 = conv_bn_layer(input, ch_out, 3, stride, 1)
conv2 = conv_bn_layer(conv1, ch_out, 3, 1, 1, act=None)
return fluid.layers.elementwise_add(x=short, y=conv2, act='relu')
def bottleneck(input, ch_out, stride):
short = shortcut(input, ch_out * 4, stride)
conv1 = conv_bn_layer(input, ch_out, 1, stride, 0)
conv2 = conv_bn_layer(conv1, ch_out, 3, 1, 1)
conv3 = conv_bn_layer(conv2, ch_out * 4, 1, 1, 0, act=None)
return fluid.layers.elementwise_add(x=short, y=conv3, act='relu')
def layer_warp(block_func, input, ch_out, count, stride):
res_out = block_func(input, ch_out, stride)
for i in range(1, count):
res_out = block_func(res_out, ch_out, 1)
return res_out
def resnet_imagenet(input, class_dim, depth=50, data_format='NCHW'):
cfg = {
18: ([2, 2, 2, 1], basicblock),
34: ([3, 4, 6, 3], basicblock),
50: ([3, 4, 6, 3], bottleneck),
101: ([3, 4, 23, 3], bottleneck),
152: ([3, 8, 36, 3], bottleneck)
}
stages, block_func = cfg[depth]
conv1 = conv_bn_layer(input, ch_out=64, filter_size=7, stride=2, padding=3)
pool1 = fluid.layers.pool2d(
input=conv1, pool_type='avg', pool_size=3, pool_stride=2)
res1 = layer_warp(block_func, pool1, 64, stages[0], 1)
res2 = layer_warp(block_func, res1, 128, stages[1], 2)
res3 = layer_warp(block_func, res2, 256, stages[2], 2)
res4 = layer_warp(block_func, res3, 512, stages[3], 2)
pool2 = fluid.layers.pool2d(
input=res4,
pool_size=7,
pool_type='avg',
pool_stride=1,
global_pooling=True)
out = fluid.layers.fc(input=pool2, size=class_dim, act='softmax')
return out
def resnet_cifar10(input, class_dim, depth=32, data_format='NCHW'):
assert (depth - 2) % 6 == 0
n = (depth - 2) // 6
conv1 = conv_bn_layer(
input=input, ch_out=16, filter_size=3, stride=1, padding=1)
res1 = layer_warp(basicblock, conv1, 16, n, 1)
res2 = layer_warp(basicblock, res1, 32, n, 2)
res3 = layer_warp(basicblock, res2, 64, n, 2)
pool = fluid.layers.pool2d(
input=res3, pool_size=8, pool_type='avg', pool_stride=1)
out = fluid.layers.fc(input=pool, size=class_dim, act='softmax')
return out
def get_model(args):
model = resnet_cifar10
if args.data_set == "cifar10":
class_dim = 10
if args.data_format == 'NCHW':
dshape = [3, 32, 32]
else:
dshape = [32, 32, 3]
model = resnet_cifar10
train_reader = paddle.dataset.cifar.train10()
test_reader = paddle.dataset.cifar.test10()
elif args.data_set == "flowers":
class_dim = 102
if args.data_format == 'NCHW':
dshape = [3, 224, 224]
else:
dshape = [224, 224, 3]
model = resnet_imagenet
train_reader = paddle.dataset.flowers.train()
test_reader = paddle.dataset.flowers.test()
elif args.data_set == "imagenet":
class_dim = 1000
if args.data_format == 'NCHW':
dshape = [3, 224, 224]
else:
dshape = [224, 224, 3]
model = resnet_imagenet
if not args.data_path:
raise Exception(
"Must specify --data_path when training with imagenet")
train_reader = imagenet_train(args.data_path)
test_reader = imagenet_test(args.data_path)
if args.use_reader_op:
filelist = [
os.path.join(args.data_path, f) for f in os.listdir(args.data_path)
]
data_file = fluid.layers.open_files(
filenames=filelist,
shapes=[[-1] + dshape, (-1, 1)],
lod_levels=[0, 0],
dtypes=["float32", "int64"],
thread_num=args.gpus,
pass_num=args.pass_num)
data_file = fluid.layers.double_buffer(
fluid.layers.batch(
data_file, batch_size=args.batch_size))
input, label = fluid.layers.read_file(data_file)
else:
input = fluid.layers.data(name='data', shape=dshape, dtype='float32')
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
if args.device == 'CPU' and args.cpus > 1:
places = fluid.layers.get_places(args.cpus)
pd = fluid.layers.ParallelDo(places)
with pd.do():
predict = model(pd.read_input(input), class_dim)
label = pd.read_input(label)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
batch_acc = fluid.layers.accuracy(input=predict, label=label)
pd.write_output(avg_cost)
pd.write_output(batch_acc)
avg_cost, batch_acc = pd()
avg_cost = fluid.layers.mean(avg_cost)
batch_acc = fluid.layers.mean(batch_acc)
else:
predict = model(input, class_dim)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
batch_acc = fluid.layers.accuracy(input=predict, label=label)
inference_program = fluid.default_main_program().clone()
with fluid.program_guard(inference_program):
inference_program = fluid.io.get_inference_program(
target_vars=[batch_acc])
optimizer = fluid.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
batched_train_reader = paddle.batch(
train_reader if args.no_random else paddle.reader.shuffle(
train_reader, buf_size=5120),
batch_size=args.batch_size * args.gpus,
drop_last=True)
batched_test_reader = paddle.batch(
test_reader, batch_size=args.batch_size, drop_last=True)
return avg_cost, inference_program, optimizer, batched_train_reader,\
batched_test_reader, batch_acc
benchmark/fluid/models/stacked_dynamic_lstm.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import cPickle
import os
import random
import time
import numpy
import paddle
import paddle.dataset.imdb as imdb
import paddle.fluid as fluid
import paddle.batch as batch
import paddle.fluid.profiler as profiler
word_dict = imdb.word_dict()
def crop_sentence(reader, crop_size):
unk_value = word_dict['<unk>']
def __impl__():
for item in reader():
if len([x for x in item[0] if x != unk_value]) < crop_size:
yield item
return __impl__
def get_model(args):
if args.use_reader_op:
raise Exception(
"stacked_dynamic_lstm do not support reader op for now.")
lstm_size = 512
emb_dim = 512
crop_size = 1500
data = fluid.layers.data(
name="words", shape=[1], lod_level=1, dtype='int64')
sentence = fluid.layers.embedding(
input=data, size=[len(word_dict), emb_dim])
sentence = fluid.layers.fc(input=sentence, size=lstm_size, act='tanh')
rnn = fluid.layers.DynamicRNN()
with rnn.block():
word = rnn.step_input(sentence)
prev_hidden = rnn.memory(value=0.0, shape=[lstm_size])
prev_cell = rnn.memory(value=0.0, shape=[lstm_size])
def gate_common(
ipt,
hidden,
size, ):
gate0 = fluid.layers.fc(input=ipt, size=size, bias_attr=True)
gate1 = fluid.layers.fc(input=hidden, size=size, bias_attr=False)
gate = fluid.layers.sums(input=[gate0, gate1])
return gate
forget_gate = fluid.layers.sigmoid(
x=gate_common(word, prev_hidden, lstm_size))
input_gate = fluid.layers.sigmoid(
x=gate_common(word, prev_hidden, lstm_size))
output_gate = fluid.layers.sigmoid(
x=gate_common(word, prev_hidden, lstm_size))
cell_gate = fluid.layers.tanh(
x=gate_common(word, prev_hidden, lstm_size))
cell = fluid.layers.sums(input=[
fluid.layers.elementwise_mul(
x=forget_gate, y=prev_cell), fluid.layers.elementwise_mul(
x=input_gate, y=cell_gate)
])
hidden = fluid.layers.elementwise_mul(
x=output_gate, y=fluid.layers.tanh(x=cell))
rnn.update_memory(prev_cell, cell)
rnn.update_memory(prev_hidden, hidden)
rnn.output(hidden)
last = fluid.layers.sequence_pool(rnn(), 'last')
logit = fluid.layers.fc(input=last, size=2, act='softmax')
loss = fluid.layers.cross_entropy(
input=logit,
label=fluid.layers.data(
name='label', shape=[1], dtype='int64'))
loss = fluid.layers.mean(x=loss)
# add acc
batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(input=logit, label=fluid.layers.data(name='label', \
shape=[1], dtype='int64'), total=batch_size_tensor)
inference_program = fluid.default_main_program().clone()
with fluid.program_guard(inference_program):
inference_program = fluid.io.get_inference_program(
target_vars=[batch_acc, batch_size_tensor])
adam = fluid.optimizer.Adam()
train_reader = batch(
paddle.reader.shuffle(
crop_sentence(imdb.train(word_dict), crop_size), buf_size=25000),
batch_size=args.batch_size * args.gpus)
test_reader = batch(
paddle.reader.shuffle(
crop_sentence(imdb.test(word_dict), crop_size), buf_size=25000),
batch_size=args.batch_size)
return loss, inference_program, adam, train_reader, test_reader, batch_acc
benchmark/fluid/models/vgg.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""VGG16 benchmark in Fluid"""
from __future__ import print_function
import sys
import time
import numpy as np
import paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import argparse
import functools
import os
def vgg16_bn_drop(input):
def conv_block(input, num_filter, groups, dropouts):
return fluid.nets.img_conv_group(
input=input,
pool_size=2,
pool_stride=2,
conv_num_filter=[num_filter] * groups,
conv_filter_size=3,
conv_act='relu',
conv_with_batchnorm=True,
conv_batchnorm_drop_rate=dropouts,
pool_type='max')
conv1 = conv_block(input, 64, 2, [0.3, 0])
conv2 = conv_block(conv1, 128, 2, [0.4, 0])
conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0])
conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0])
conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0])
drop = fluid.layers.dropout(x=conv5, dropout_prob=0.5)
fc1 = fluid.layers.fc(input=drop, size=512, act=None)
bn = fluid.layers.batch_norm(input=fc1, act='relu')
drop2 = fluid.layers.dropout(x=bn, dropout_prob=0.5)
fc2 = fluid.layers.fc(input=drop2, size=512, act=None)
return fc2
def get_model(args):
if args.data_set == "cifar10":
classdim = 10
if args.data_format == 'NCHW':
data_shape = [3, 32, 32]
else:
data_shape = [32, 32, 3]
else:
classdim = 102
if args.data_format == 'NCHW':
data_shape = [3, 224, 224]
else:
data_shape = [224, 224, 3]
if args.use_reader_op:
filelist = [
os.path.join(args.data_path, f) for f in os.listdir(args.data_path)
]
data_file = fluid.layers.open_files(
filenames=filelist,
shapes=[[-1] + data_shape, (-1, 1)],
lod_levels=[0, 0],
dtypes=["float32", "int64"],
thread_num=args.gpus,
pass_num=args.pass_num)
data_file = fluid.layers.double_buffer(
fluid.layers.batch(
data_file, batch_size=args.batch_size))
images, label = fluid.layers.read_file(data_file)
else:
images = fluid.layers.data(
name='data', shape=data_shape, dtype='float32')
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
# Train program
net = vgg16_bn_drop(images)
predict = fluid.layers.fc(input=net, size=classdim, act='softmax')
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
# Evaluator
batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(
input=predict, label=label, total=batch_size_tensor)
# inference program
inference_program = fluid.default_main_program().clone()
with fluid.program_guard(inference_program):
inference_program = fluid.io.get_inference_program(
target_vars=[batch_acc, batch_size_tensor])
# Optimization
optimizer = fluid.optimizer.Adam(learning_rate=args.learning_rate)
# data reader
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.train10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.train(),
buf_size=5120),
batch_size=args.batch_size * args.gpus)
test_reader = paddle.batch(
paddle.dataset.cifar.test10()
if args.data_set == 'cifar10' else paddle.dataset.flowers.test(),
batch_size=args.batch_size)
return avg_cost, inference_program, optimizer, train_reader, test_reader, batch_acc
benchmark/fluid/recordio_converter.py 0 → 100644
浏览文件 @ a38a8db9
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import random
import paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
from paddle.dataset import mnist, cifar, flowers, image
def convert_2_recordio(py_reader, outfilepath, batch_size, shape_data,
shape_label):
num_batches = 0
with fluid.program_guard(fluid.Program(), fluid.Program()):
reader = paddle.batch(py_reader(), batch_size=batch_size)
feeder = fluid.DataFeeder(
feed_list=[ # order is image and label
fluid.layers.data(
name='image', shape=shape_data),
fluid.layers.data(
name='label', shape=shape_label, dtype='int64'),
],
place=fluid.CPUPlace())
num_batches = fluid.recordio_writer.convert_reader_to_recordio_file(
outfilepath, reader, feeder)
return num_batches
def prepare_mnist(outpath, batch_size):
outfilepath = os.path.join(outpath, "mnist.recordio")
convert_2_recordio(mnist.train, outfilepath, batch_size, [784], [1])
def prepare_cifar10(outpath, batch_size):
outfilepath = os.path.join(outpath, "cifar.recordio")
convert_2_recordio(cifar.train10, outfilepath, batch_size, [3, 32, 32], [1])
def prepare_flowers(outpath, batch_size):
outfilepath = os.path.join(outpath, "flowers.recordio")
convert_2_recordio(flowers.train, outfilepath, batch_size, [3, 224, 224],
[1])
def default_mapper(sample):
img, label = sample
img = image.simple_transform(
img, 256, 224, True, mean=[103.94, 116.78, 123.68])
return img.flatten().astype('float32'), label
def imagenet_train(data_dir):
contents = os.listdir(data_dir)
if set(contents) != set(
["train", "train.txt", "val", "val_set", "val.txt", "unzip.sh"]):
raise Exception("Imagenet data contents error!")
img2label = dict()
imgfilelist = []
with open(os.path.join(data_dir, "train.txt")) as fn:
while 1:
l = fn.readline()
if not l:
break
img, lbl = l[:-1].split(" ")
img2label[img] = int(lbl)
imgfilelist.append(img)
# shuffle all, this is slow
random.shuffle(imgfilelist)
def train_reader():
for idx, imgfile in enumerate(imgfilelist):
data = image.load_image(
os.path.join(data_dir, "train", imgfile.lower()))
label = [img2label[imgfile], ]
yield [data, label]
return paddle.reader.map_readers(default_mapper, train_reader)
def imagenet_test(data_dir):
contents = os.listdir(data_dir)
if set(contents) != set(
["train", "train.txt", "val", "val_set", "val.txt", "unzip.sh"]):
raise Exception("Imagenet data contents error!")
img2label = dict()
imgfilelist = []
with open(os.path.join(data_dir, "val.txt")) as fn:
while 1:
l = fn.readline()
if not l:
break
img, lbl = l[:-1].split(" ")
img2label[img] = int(lbl)
imgfilelist.append(img)
def test_reader():
for idx, imgfile in enumerate(imgfilelist):
base_path = os.path.join(data_dir, "val", imgfile.split(".")[0])
image_path = ".".join([base_path, "jpeg"])
data = image.load_image(image_path)
label = [img2label[imgfile], ]
yield [data, label]
return paddle.reader.map_readers(default_mapper, test_reader)
# FIXME(wuyi): delete this when https://github.com/PaddlePaddle/Paddle/pull/11066 is merged
def convert_reader_to_recordio_files(
filename,
batch_per_file,
reader_creator,
feeder,
compressor=core.RecordIOWriter.Compressor.Snappy,
max_num_records=1000,
feed_order=None):
if feed_order is None:
feed_order = feeder.feed_names
f_name, f_ext = os.path.splitext(filename)
assert (f_ext == ".recordio")
lines = []
f_idx = 0
counter = 0
for idx, batch in enumerate(reader_creator()):
lines.append(batch)
if idx >= batch_per_file and idx % batch_per_file == 0:
filename = "%s-%05d%s" % (f_name, f_idx, f_ext)
with fluid.recordio_writer.create_recordio_writer(
filename, compressor, max_num_records) as writer:
for l in lines:
res = feeder.feed(l)
for each in feed_order:
writer.append_tensor(res[each])
writer.complete_append_tensor()
counter += 1
lines = []
f_idx += 1
print("written file: ", filename)
return counter
def prepare_imagenet(inpath, outpath, batch_size):
r = paddle.batch(imagenet_train(inpath), batch_size=batch_size)
feeder = fluid.DataFeeder(
feed_list=[
fluid.layers.data(
name="image", shape=[3, 224, 224]), fluid.layers.data(
name="label", shape=[1], dtype='int64')
],
place=fluid.CPUPlace())
outpath = os.path.join(outpath, "imagenet.recordio")
convert_reader_to_recordio_files(outpath, 10000, r, feeder)
benchmark/fluid/resnet.py 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
benchmark/fluid/run.sh
浏览文件 @ a38a8db9
......@@ -2,6 +2,7 @@
# This script benchmarking the PaddlePaddle Fluid on
# single thread single GPU.
mkdir -p logs
#export FLAGS_fraction_of_gpu_memory_to_use=0.0
export CUDNN_PATH=/paddle/cudnn_v5
......@@ -35,71 +36,74 @@ nohup stdbuf -oL nvidia-smi \
--format=csv \
--filename=mem.log \
-l 1 &
# mnist
# mnist gpu mnist 128
FLAGS_benchmark=true stdbuf -oL python fluid/mnist.py \
FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
--model=mnist \
--device=GPU \
--batch_size=128 \
--skip_batch_num=5 \
--iterations=500 \
2>&1 | tee -a mnist_gpu_128.log
2>&1 | tee -a logs/mnist_gpu_128.log
# vgg16
# gpu cifar10 128
FLAGS_benchmark=true stdbuf -oL python fluid/vgg16.py \
FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
--model=vgg16 \
--device=GPU \
--batch_size=128 \
--skip_batch_num=5 \
--iterations=30 \
2>&1 | tee -a vgg16_gpu_128.log
2>&1 | tee -a logs/vgg16_gpu_128.log
# flowers gpu 128
FLAGS_benchmark=true stdbuf -oL python fluid/vgg16.py \
FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
--model=vgg16 \
--device=GPU \
--batch_size=32 \
--data_set=flowers \
--skip_batch_num=5 \
--iterations=30 \
2>&1 | tee -a vgg16_gpu_flowers_32.log
2>&1 | tee -a logs/vgg16_gpu_flowers_32.log
# resnet50
# resnet50 gpu cifar10 128
FLAGS_benchmark=true stdbuf -oL python fluid/resnet50.py \
FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
--model=resnet \
--device=GPU \
--batch_size=128 \
--data_set=cifar10 \
--model=resnet_cifar10 \
--skip_batch_num=5 \
--iterations=30 \
2>&1 | tee -a resnet50_gpu_128.log
2>&1 | tee -a logs/resnet50_gpu_128.log
# resnet50 gpu flowers 64
FLAGS_benchmark=true stdbuf -oL python fluid/resnet50.py \
FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
--model=resnet \
--device=GPU \
--batch_size=64 \
--data_set=flowers \
--model=resnet_imagenet \
--skip_batch_num=5 \
--iterations=30 \
2>&1 | tee -a resnet50_gpu_flowers_64.log
2>&1 | tee -a logs/resnet50_gpu_flowers_64.log
# lstm
# lstm gpu imdb 32 # tensorflow only support batch=32
FLAGS_benchmark=true stdbuf -oL python fluid/stacked_dynamic_lstm.py \
FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
--model=stacked_dynamic_lstm \
--device=GPU \
--batch_size=32 \
--skip_batch_num=5 \
--iterations=30 \
--hidden_dim=512 \
--emb_dim=512 \
--crop_size=1500 \
2>&1 | tee -a lstm_gpu_32.log
2>&1 | tee -a logs/lstm_gpu_32.log
# seq2seq
# seq2seq gpu wmb 128
FLAGS_benchmark=true stdbuf -oL python fluid/machine_translation.py \
FLAGS_benchmark=true stdbuf -oL python fluid_benchmark.py \
--model=machine_translation \
--device=GPU \
--batch_size=128 \
--skip_batch_num=5 \
--iterations=30 \
2>&1 | tee -a lstm_gpu_128.log
2>&1 | tee -a logs/lstm_gpu_128.log
benchmark/fluid/run_fluid_benchmark.sh 0 → 100644
浏览文件 @ a38a8db9
#!/bin/bash
PADDLE_TRAINING_ROLE=PSERVER PADDLE_PSERVER_PORT=7164 PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=2 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=0 python fluid_benchmark.py --model resnet --device CPU --update_method pserver --iterations=10000 &
sleep 15
CUDA_VISIBLE_DEVICES=0,1 PADDLE_TRAINING_ROLE=TRAINER PADDLE_PSERVER_PORT=7164 PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=2 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=0 python fluid_benchmark.py --model resnet --device GPU --update_method pserver --iterations=10000 --gpus 2 &
CUDA_VISIBLE_DEVICES=2,3 PADDLE_TRAINING_ROLE=TRAINER PADDLE_PSERVER_PORT=7164 PADDLE_PSERVER_IPS=127.0.0.1 PADDLE_TRAINERS=2 PADDLE_CURRENT_IP=127.0.0.1 PADDLE_TRAINER_ID=1 python fluid_benchmark.py --model resnet --device GPU --update_method pserver --iterations=10000 --gpus 2 &
benchmark/fluid/stacked_dynamic_lstm.py 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
benchmark/fluid/vgg.py 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
benchmark/paddle/image/run.sh
浏览文件 @ a38a8db9
#!/bin/bash
set -e
function train() {
......
benchmark/paddle/image/run_mkl_infer.sh
浏览文件 @ a38a8db9
#!/bin/bash
set -e
function clock_to_seconds() {
......
benchmark/paddle/image/run_mkl_train.sh
浏览文件 @ a38a8db9
#!/bin/bash
set -e
function train() {
......
benchmark/paddle/image/run_openblas_infer.sh
浏览文件 @ a38a8db9
#!/bin/bash
set -e
function clock_to_seconds() {
......
benchmark/paddle/rnn/run.sh
浏览文件 @ a38a8db9
#!/bin/bash
set -e
function train() {
......
benchmark/tensorflow/image/run.sh
浏览文件 @ a38a8db9
#!/bin/bash
set -e
function test() {
......
benchmark/tensorflow/image/run_multi.sh
浏览文件 @ a38a8db9
#!/bin/bash
set -e
function test() {
......
benchmark/tensorflow/rnn/run.sh
浏览文件 @ a38a8db9
#!/bin/bash
set -e
function test() {
......
cmake/cblas.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/configure.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/cpplint.cmake 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
cmake/cudnn.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/anakin.cmake 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/boost.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/brpc.cmake 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/cub.cmake 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/eigen.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/grpc.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/leveldb.cmake 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/libxsmm.cmake 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/mkldnn.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/mklml.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/openblas.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/protobuf.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/python.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/snappy.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/snappystream.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/external/xbyak.cmake 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
cmake/generic.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/inference_lib.cmake
浏览文件 @ a38a8db9
此差异已折叠。
cmake/version.cmake
浏览文件 @ a38a8db9
此差异已折叠。
contrib/float16/README.md 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
contrib/float16/float16_transpiler.py 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
contrib/float16/run_float16_demo.sh 已删除 100755 → 0
浏览文件 @ 10ec329b
此差异已折叠。
doc/about/about_us.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/CMakeLists.txt
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/CMakeLists.txt
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/average.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/backward.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/clip.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/data.rst 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
doc/fluid/api/data_feeder.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/evaluator.rst 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
doc/fluid/api/executor.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/fluid.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/gen_doc.py
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/gen_doc.sh
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/index_en.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/initializer.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/io.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/layers.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/metrics.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/nets.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/optimizer.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/param_attr.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/profiler.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/recordio_writer.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/regularizer.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/api/transpiler.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/design/concepts/python_data_feeding.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/design/dist_train/dist_train_nccl2.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/design/dist_train/src/ncc2_design.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/design/dist_train/src/ncc2_design.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/design/ir/overview.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/design/motivation/api.md
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/design/quantization/quantization_forward.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/dev/api_doc_std_cn.md
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/dev/api_doc_std_en.md
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/getstarted/index_cn.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/getstarted/index_en.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/howto/cluster/fluid_recordio.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/howto/cluster/nccl2_rdma_training.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/howto/index_cn.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/howto/inference/build_and_install_lib_cn.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/howto/inference/index_cn.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/howto/optimization/timeline.md 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
doc/fluid/howto/optimization/timeline_cn.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/howto/optimization/timeline_en.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/1.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/2.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/3.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/4.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/LoDTensor.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/compile_run_time.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/executor.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/fluid_examples.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/fluid_module_1.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/fluid_module_2.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/layer.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/operator1.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/operator2.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/place.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/print_fluid_program.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/program_desc1.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/program_desc2.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/raw_input.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/scope_variable_tensor.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/sorted_input.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/transpiler.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/images/user_interface.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/index_cn.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/benchmark.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/deploy/mobile_dev.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/development/new_op.md 0 → 120000
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/development/nvvp1.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/development/nvvp2.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/development/nvvp3.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/development/nvvp4.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/development/pprof_1.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/development/pprof_2.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/index.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/advanced_usage/pics/anakin_fm_ch.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/beginners_guide/basics/index.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/beginners_guide/index.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/beginners_guide/install/paddleci.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/faq/faq.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/faq/index_cn.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/user_guides/howto/debug/index.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/user_guides/howto/debug/visualdl.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/user_guides/howto/training/index.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/user_guides/index.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/fluid/new_docs/user_guides/models/index.rst 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/mobile/CMakeLists.txt
浏览文件 @ a38a8db9
此差异已折叠。
doc/mobile/index_cn.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/survey/dynamic_graph.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/survey/op_fusion_design.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/templates/conf.py.cn.in
浏览文件 @ a38a8db9
此差异已折叠。
doc/templates/conf.py.en.in
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/CMakeLists.txt
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/api/CMakeLists.txt
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/api/config/evaluators.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/api/config/layer.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/api/index_en.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/design/mkl/mkldnn.md
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/dev/new_layer_cn.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/dev/new_layer_en.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/faq/parameter/index_cn.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/faq/parameter/index_en.rst
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/FullyConnected.jpg 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/add_security_group.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/bi_lstm.jpg 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/checkpointing.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/create_efs.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/csr.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/data_dispatch.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/dataset.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/dataset.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/doc_en.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/efs_mount.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/encoder-decoder-attention-model.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/engine.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/file_storage.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/file_storage.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/glossary_rnn.dot 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/glossary_rnn_with_memory.dot 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/gradients.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/init_lock.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/init_lock.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/k8s-paddle-arch.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/layers.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/managed_policy.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/matrix.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/nvvp1.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/nvvp2.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/nvvp3.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/nvvp4.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/overview.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-cloud-in-data-center.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-etcd.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-etcd.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-model-sharding.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-model-sharding.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-ps-0.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-ps-1.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-ps.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-task-queues.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-task-queues.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-task-states.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/paddle-task-states.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/ps_cn.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/ps_en.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/pserver_and_trainer.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/pserver_init.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/pserver_init.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/route53_create_recordset.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/route53_create_zone.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/sequence_data.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/simple_full_hierarchical_recurrent.dot 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/simple_full_recurrent.dot 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/submit-job.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/submit-job.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/trainer.graffle 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/trainer.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/trainer_cn.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/worker_security_group.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
doc/v2/images/workflow_of_CAPI.png 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
go/pserver/optimizer.go
浏览文件 @ a38a8db9
此差异已折叠。
paddle/.gitignore
浏览文件 @ a38a8db9
此差异已折叠。
paddle/CMakeLists.txt
浏览文件 @ a38a8db9
此差异已折叠。
paddle/api/Arguments.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/ConfigParser.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/GradientMachine.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/Matrix.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/Paddle.i 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/PaddleAPI.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/PaddleAPIPrivate.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/Parameter.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/ParameterOptimizer.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/ParameterUpdater.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/SequenceGenerator.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/Trainer.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/Util.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/api/Vector.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/capi/Arguments.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/capi/CMakeLists.txt 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/capi/Main.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/capi/arguments.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/capi/capi_private.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/capi/gradient_machine.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/capi/tests/test_Arguments.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/capi/tests/test_GradientMachine.cpp 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/contrib/float16/README.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/contrib/float16/float16_transpiler.py 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/contrib/float16/run_float16_demo.sh 0 → 100755
浏览文件 @ a38a8db9
此差异已折叠。
paddle/cuda/CMakeLists.txt 已删除 100755 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/include/hl_activation_functions.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/include/hl_base.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/include/hl_gpu_gru.cuh 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/include/hl_gpu_lstm.cuh 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/include/hl_gpu_matrix_kernel.cuh 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/include/hl_tensor_ops.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_cuda_aggregate.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_cuda_cublas.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_cuda_cudnn.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_cuda_device.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_cuda_lstm.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_cuda_matrix.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_cuda_sequence.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_cuda_sparse.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_table_apply.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_top_k.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/cuda/src/hl_warpctc_wrap.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/API.spec 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/CMakeLists.txt
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/block_desc.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/data_type.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/data_type.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/data_type_test.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/all_reduce_op_handle.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/all_reduce_op_handle.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/build_strategy.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/data_balance_op_handle.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/exception_holder.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/execution_strategy.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/fuse_vars_op_handle.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/fuse_vars_op_handle.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/multi_devices_helper.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/multi_devices_helper.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/rpc_op_handle.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/rpc_op_handle.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/details/ssa_graph.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/framework/details/ssa_graph.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/framework/executor.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/executor.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/init.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/framework/init_test.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/framework/ir/CMakeLists.txt 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/graph.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/graph.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/graph_helper.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/graph_helper.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/graph_helper_test.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/graph_test.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/graph_viz_pass.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/graph_viz_pass.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/node.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/node.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/pass.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/pass.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/ir/pass_test.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/lod_tensor.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/mixed_vector_test.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/op_desc.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/op_desc.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/op_info.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/operator.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/operator.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/reader.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/reader.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/reader_test.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/scope.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/scope.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/tensor.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/tensor.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/type_defs.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/framework/var_type.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/CMakeLists.txt 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/README.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/analyzer.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/analyzer.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/analyzer_main.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/analyzer_tester.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/argument.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/argument.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/data_flow_graph.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/data_flow_graph.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/device.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/dot.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/dot.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/dot_tester.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/graph_traits.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/graph_traits.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/helper.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/helper.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/model_store_pass.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/model_store_pass.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/node.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/node.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/node_attr_flags.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/node_tester.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/pass.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/pass.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/pass_manager.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/pass_manager.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/pass_manager_tester.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/subgraph_splitter.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/subgraph_splitter.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/analysis/ut_helper.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/CMakeLists.txt 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/README.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/api.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/api_anakin_engine.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/api_anakin_engine.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/api_anakin_engine_tester.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/api_impl.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/api_impl.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/api_impl_tester.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/api_tester.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/demo_ci/.gitignore 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/demo_ci/CMakeLists.txt 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/demo_ci/README.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/demo_ci/clean.sh 0 → 100755
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/demo_ci/run.sh 0 → 100755
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/demo_ci/utils.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/demo_ci/vis_demo.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/high_level_api.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/high_level_api_cn.md 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/api/paddle_inference_api.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/check_symbol.sh 0 → 100755
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/engine.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/io.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/io.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/paddle_fluid.sym 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/tensorrt/convert/fc_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/tensorrt/convert/pool2d_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/tensorrt/convert/softmax_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/tensorrt/convert/test_fc_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/tensorrt/convert/test_mul_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/tensorrt/convert/ut_helper.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/inference/tests/book/test_inference_nlp.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/memory/malloc.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/.flatten_op.cc.swp 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/adam_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/adam_op.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/adamax_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/arg_max_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/arg_max_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/arg_min_max_op_base.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/arg_min_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/arg_min_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/argsort_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/argsort_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/argsort_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/assign_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/auc_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/auc_op.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/bipartite_match_op.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/box_coder_op.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/box_coder_op.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/box_coder_op.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/cast_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/cast_op.cu
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/checkpoint_notify_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/clip_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/concat_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/concat_op.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/conv_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/conv_op.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/crop_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/crop_op.cu
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/crop_op.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/cumsum_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detail/CMakeLists.txt 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/detail/grpc_client.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/detail/grpc_client.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/detail/grpc_server.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/detail/grpc_server.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/detail/grpc_service.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/detail/macros.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detail/send_recv.proto 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/detail/serde_test.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/detection/CMakeLists.txt 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/anchor_generator_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/anchor_generator_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/anchor_generator_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/bipartite_match_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/box_coder_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/box_coder_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/box_coder_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/iou_similarity_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/iou_similarity_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/multiclass_nms_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/prior_box_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/prior_box_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/prior_box_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/target_assign_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/target_assign_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/detection/target_assign_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/CMakeLists.txt 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/brpc_client.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/brpc_client.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/brpc_server.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/brpc_server.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/grpc_client.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/grpc_client.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/grpc_serde.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/grpc_serde.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/grpc_serde_test.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/grpc_server.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/grpc_server.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/grpc_service.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/request_handler.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/rpc_client.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/rpc_client.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/rpc_server.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/rpc_server.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/rpc_server_test.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/send_recv.proto.in 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/sendrecvop_utils.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/sendrecvop_utils.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/variable_response.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/distributed/variable_response.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/elementwise_add_mkldnn_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/expand_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/extract_rows_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fake_dequantize_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fake_dequantize_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fake_dequantize_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fake_quantize_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fake_quantize_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fake_quantize_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fc_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fc_op.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/feed_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fetch_barrier_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fetch_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fill_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/flatten_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/ftrl_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fused_elemwise_activation_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fused_elemwise_activation_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/fused_elemwise_activation_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/gather_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/gaussian_random_mkldnn_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/gen_nccl_id_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/go_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/gru_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/hierarchical_sigmoid_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/hierarchical_sigmoid_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/iou_similarity_op.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/iou_similarity_op.cu 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/is_empty_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/load_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/lrn_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/lstm_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/lstmp_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/math/blas.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/math/blas.h
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/math/functors.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/math/im2col_cfo_cpu.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/math/matmul.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/math/matrix_bit_code.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/math/matrix_bit_code.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/matmul_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/matmul_op.cu.cc 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/matmul_op.h 已删除 100644 → 0
浏览文件 @ 10ec329b
此差异已折叠。
paddle/fluid/operators/maxout_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/mean_iou_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/mean_iou_op.cu 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/mean_iou_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/mean_op.cc
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/merge_ids_op.cc 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/merge_ids_op.h 0 → 100644
浏览文件 @ a38a8db9
此差异已折叠。
paddle/fluid/operators/minus_op.cc
浏览文件 @ a38a8db9
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