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 d2ba05a6 编写于 作者: T tangwei12
浏览文件
操作

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

上级 10669f1f a79676e8
展开全部 隐藏空白更改
内联 并排
Showing with 28691 addition and 7427 deletion
.gitignore
浏览文件 @ d2ba05a6
......@@ -25,12 +25,3 @@ third_party/
# clion workspace.
cmake-build-*
# generated while compiling
paddle/pybind/pybind.h
CMakeFiles
cmake_install.cmake
paddle/.timestamp
python/paddlepaddle.egg-info/
paddle/fluid/pybind/pybind.h
python/paddle/version.py
.pre-commit-config.yaml
浏览文件 @ d2ba05a6
repos:
- repo: https://github.com/Lucas-C/pre-commit-hooks.git
sha: v1.0.1
hooks:
......@@ -25,6 +26,14 @@
entry: bash ./.clang_format.hook -i
language: system
files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|proto)$
- repo: local
hooks:
- id: cpplint-cpp-source
name: cpplint
description: Check C++ code style using cpplint.py.
entry: bash ./tools/codestyle/cpplint_pre_commit.hook
language: system
files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx)$
- repo: https://github.com/PaddlePaddle/pre-commit-golang
sha: 8337620115c25ff8333f1b1a493bd031049bd7c0
hooks:
......
.travis.yml
浏览文件 @ d2ba05a6
......@@ -34,7 +34,7 @@ addons:
- automake
- libtool
- ccache
ssh_known_hosts: 52.76.173.135
ssh_known_hosts: 13.229.163.131
before_install:
- if [[ "$JOB" == "check_style" ]]; then sudo ln -s /usr/bin/clang-format-3.8 /usr/bin/clang-format; fi
# Paddle is using protobuf 3.1 currently. Protobuf 3.2 breaks the compatibility. So we specify the python
......
CMakeLists.txt
浏览文件 @ d2ba05a6
......@@ -39,6 +39,7 @@ option(WITH_GPU "Compile PaddlePaddle with NVIDIA GPU" ${CUDA_F
option(WITH_AMD_GPU "Compile PaddlePaddle with AMD GPU" OFF)
option(WITH_AVX "Compile PaddlePaddle with AVX intrinsics" ${AVX_FOUND})
option(WITH_MKL "Compile PaddlePaddle with MKL support." ${AVX_FOUND})
option(WITH_TENSORRT "Compile PaddlePaddle with TensorRT support." OFF)
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)
......@@ -53,8 +54,7 @@ option(WITH_COVERAGE "Compile PaddlePaddle with code coverage" OFF)
option(COVERALLS_UPLOAD "Package code coverage data to coveralls" OFF)
option(ON_TRAVIS "Exclude special unit test on Travis CI" OFF)
option(WITH_C_API "Compile PaddlePaddle with C-API(Prediction)" OFF)
# TODO: Only compile PaddlePaddle fluid version by WITH_FLUID option.
option(WITH_FLUID "Compile PaddlePaddle fluid only(TODO)" OFF)
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)
......@@ -109,7 +109,7 @@ if (WITH_C_API AND WITH_PYTHON)
endif()
if (WITH_C_API)
set(WITH_FLUID OFF CACHE STRING "Disable install fluid when compile the C_API" FORCE)
set(WITH_FLUID_ONLY OFF CACHE STRING "Disable install fluid when compile the C_API" FORCE)
endif()
if(MOBILE_INFERENCE)
......@@ -147,6 +147,7 @@ include(external/cares)
include(external/grpc)
include(external/snappy) # download snappy
include(external/snappystream)
include(external/threadpool)
include(cudnn) # set cudnn libraries, must before configure
include(cupti)
......@@ -181,6 +182,11 @@ if(WITH_GPU)
include(cuda)
endif(WITH_GPU)
# TensorRT depends on GPU.
if (NOT WITH_GPU)
set(WITH_TENSORRT OFF)
endif()
if(WITH_AMD_GPU)
find_package(HIP)
include(hip)
......
Dockerfile
浏览文件 @ d2ba05a6
# A image for building paddle binaries
# Use cuda devel base image for both cpu and gpu environment
FROM nvidia/cuda:8.0-cudnn5-devel-ubuntu16.04
FROM nvidia/cuda:8.0-cudnn7-devel-ubuntu16.04
MAINTAINER PaddlePaddle Authors <paddle-dev@baidu.com>
ARG UBUNTU_MIRROR
......@@ -45,6 +45,13 @@ ENV PATH=${PATH}:${GOROOT}/bin:${GOPATH}/bin
# install glide
RUN curl -s -q https://glide.sh/get | sh
# Install TensorRT
# The unnecessary files has been removed to make the library small.
RUN wget -qO- http://paddlepaddledeps.bj.bcebos.com/TensorRT-4.0.0.3.Ubuntu-16.04.4.x86_64-gnu.cuda-8.0.cudnn7.0.tar.gz | \
tar -xz -C /usr/local && \
cp -rf /usr/local/TensorRT/include /usr && \
cp -rf /usr/local/TensorRT/lib /usr
# git credential to skip password typing
RUN git config --global credential.helper store
......@@ -57,7 +64,7 @@ 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 && \
RUN pip install --upgrade pip==9.0.3 && \
pip install -U wheel && \
pip install -U docopt PyYAML sphinx==1.5.6 && \
pip install sphinx-rtd-theme==0.1.9 recommonmark
......
benchmark/cluster/README.md
浏览文件 @ d2ba05a6
......@@ -36,11 +36,41 @@
- Trainer Count: 100
- Metrics: mini-batch / sec
| Batch Size | 32 | 64 | 128 | 256 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | - | - | - | - |
| PaddlePaddle v2 | - | - | - | - |
| TensorFlow | - | - | - | - |
<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
......@@ -48,11 +78,41 @@
- Batch Size: 64
- Metrics: mini-batch / sec
| PServer Count | 10 | 20 | 40 | 60 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | - | - | - | - |
| PaddlePaddle v2 | - | - | - | - |
| TensorFlow | - | - | - | - |
<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
......@@ -67,11 +127,69 @@ The parallel efficiency is:
$E = \div(S, N)$
| Trainer Counter | 1 | 10 | 20 | 30 | 40 | 50 | 60 | 70 | 80 | 90 | 100 |
| -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | - | - | - | - | - | - | - | - | - | - | - |
| PaddlePaddle v2 | - | - | - | - | - | - | - | - | - | - | - | - |
| TensorFlow | - | - | - | - | - | - | - | - | - | - | - | - | - |
<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
......
benchmark/cluster/vgg16/README.md
浏览文件 @ d2ba05a6
......@@ -16,11 +16,41 @@ Setting environment variable: `MKL_NUM_THREADS=1`.
- Metrics: samples / sec
| Batch Size | 32 | 64 | 128 | 256 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | 15.44 | 16.32 | 16.74 | 16.79 |
| PaddlePaddle v2 | 15.97 | 17.04 | 17.60 | 17.83 |
| TensorFlow | 9.09 | 9.10 | 9.24 | 8.66 |
<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
......@@ -28,12 +58,40 @@ Setting environment variable: `MKL_NUM_THREADS=1`.
- Trainer Count: 20
- Metrics: samples / sec
| Batch Size | 32 | 64 | 128 | 256 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | 190.20 | 222.15 | 247.40 | 258.18 |
| PaddlePaddle v2 | 170.96 | 233.71 | 256.14 | 329.23 |
| TensorFlow | - | - | - | - |
<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
......@@ -41,11 +99,41 @@ Setting environment variable: `MKL_NUM_THREADS=1`.
- Batch Size: 128
- Metrics: samples / sec
| Trainer Count | 20 | 40 | 80 | 100 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid | 263.29 (78.64%) | 518.80 (77.47%) | 836.26 (62.44%) | 1019.29 (60.89%) |
| PaddlePaddle v2 (need more tests) | 326.85 (92.85%) | 534.58 (75.93%) | 853.30 (60.60%) | 1041.99 (59.20%) |
| TensorFlow | - | - | - | - |
<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
......@@ -53,11 +141,41 @@ Setting environment variable: `MKL_NUM_THREADS=1`.
- Batch Size: 128
- Metrics: samples/ sec
| PServer Count | 3 | 6 |10 | 20 |
| -- | -- | -- | -- | -- |
| PaddlePaddle Fluid(should fix in next PR) | 589.1 | 592.6 | 656.4 | 655.8 |
| PaddlePaddle v2 | 593.4 | 791.3 | 729.7 | 821.7 |
| TensorFlow | - | - | - | - |
<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.*
......
benchmark/fluid/machine_translation.py 0 → 100644
浏览文件 @ d2ba05a6
# 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.v2 as 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 0 → 100644
浏览文件 @ d2ba05a6
# 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.v2 as 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/resnet.py 0 → 100644
浏览文件 @ d2ba05a6
# 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 functools
import numpy as np
import time
import cProfile, pstats, StringIO
import paddle.v2 as paddle
import paddle.fluid as fluid
import paddle.fluid.core as core
import paddle.fluid.profiler as profiler
def parse_args():
parser = argparse.ArgumentParser('Convolution model benchmark.')
parser.add_argument(
'--model',
type=str,
choices=['resnet_imagenet', 'resnet_cifar10'],
default='resnet_imagenet',
help='The model architecture.')
parser.add_argument(
'--batch_size', type=int, default=32, help='The minibatch size.')
parser.add_argument(
'--use_fake_data',
action='store_true',
help='use real data or fake data')
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(
'--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(
'--with_test',
action='store_true',
help='If set, test the testset during training.')
args = parser.parse_args()
return args
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 run_benchmark(model, args):
if args.use_cprof:
pr = cProfile.Profile()
pr.enable()
if args.data_set == "cifar10":
class_dim = 10
if args.data_format == 'NCHW':
dshape = [3, 32, 32]
else:
dshape = [32, 32, 3]
else:
class_dim = 102
if args.data_format == 'NCHW':
dshape = [3, 224, 224]
else:
dshape = [224, 224, 3]
input = fluid.layers.data(name='data', shape=dshape, dtype='float32')
label = fluid.layers.data(name='label', shape=[1], dtype='int64')
predict = model(input, class_dim)
cost = fluid.layers.cross_entropy(input=predict, label=label)
avg_cost = fluid.layers.mean(x=cost)
batch_size_tensor = fluid.layers.create_tensor(dtype='int64')
batch_acc = fluid.layers.accuracy(
input=predict, label=label, 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])
optimizer = fluid.optimizer.Momentum(learning_rate=0.01, momentum=0.9)
opts = optimizer.minimize(avg_cost)
fluid.memory_optimize(fluid.default_main_program())
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)
def test(exe):
test_accuracy = fluid.average.WeightedAverage()
for batch_id, data in enumerate(test_reader()):
img_data = np.array(map(lambda x: x[0].reshape(dshape),
data)).astype("float32")
y_data = np.array(map(lambda x: x[1], data)).astype("int64")
y_data = y_data.reshape([-1, 1])
acc, weight = exe.run(inference_program,
feed={"data": img_data,
"label": y_data},
fetch_list=[batch_acc, batch_size_tensor])
test_accuracy.add(value=acc, weight=weight)
return test_accuracy.eval()
place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
accuracy = fluid.average.WeightedAverage()
if args.use_fake_data:
data = train_reader().next()
image = np.array(map(lambda x: x[0].reshape(dshape), data)).astype(
'float32')
label = np.array(map(lambda x: x[1], data)).astype('int64')
label = label.reshape([-1, 1])
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
if not args.use_fake_data:
image = np.array(map(lambda x: x[0].reshape(dshape),
data)).astype('float32')
label = np.array(map(lambda x: x[1], data)).astype('int64')
label = label.reshape([-1, 1])
loss, acc, weight = exe.run(
fluid.default_main_program(),
feed={'data': image,
'label': label},
fetch_list=[avg_cost, batch_acc, batch_size_tensor])
iters += 1
num_samples += len(label)
accuracy.add(value=acc, weight=weight)
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:
pass_test_acc = test(exe)
exit(0)
def print_arguments(args):
vars(args)['use_nvprof'] = (vars(args)['use_nvprof'] and
vars(args)['device'] == 'GPU')
print('----------- resnet Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
if __name__ == '__main__':
model_map = {
'resnet_imagenet': resnet_imagenet,
'resnet_cifar10': resnet_cifar10
}
args = parse_args()
print_arguments(args)
if args.data_format == 'NHWC':
raise ValueError('Only support NCHW data_format now.')
if args.use_nvprof and args.device == 'GPU':
with profiler.cuda_profiler("cuda_profiler.txt", 'csv') as nvprof:
run_benchmark(model_map[args.model], args)
else:
run_benchmark(model_map[args.model], args)
benchmark/fluid/run.sh 0 → 100644
浏览文件 @ d2ba05a6
#!/bin/bash
# This script benchmarking the PaddlePaddle Fluid on
# single thread single GPU.
#export FLAGS_fraction_of_gpu_memory_to_use=0.0
export CUDNN_PATH=/paddle/cudnn_v5
# disable openmp and mkl parallel
#https://github.com/PaddlePaddle/Paddle/issues/7199
export MKL_NUM_THREADS=1
export OMP_NUM_THREADS=1
ht=`lscpu |grep "per core"|awk -F':' '{print $2}'|xargs`
if [ $ht -eq 1 ]; then # HT is OFF
if [ -z "$KMP_AFFINITY" ]; then
export KMP_AFFINITY="granularity=fine,compact,0,0"
fi
if [ -z "$OMP_DYNAMIC" ]; then
export OMP_DYNAMIC="FALSE"
fi
else # HT is ON
if [ -z "$KMP_AFFINITY" ]; then
export KMP_AFFINITY="granularity=fine,compact,1,0"
fi
fi
# disable multi-gpu if have more than one
export CUDA_VISIBLE_DEVICES=0
export LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$CUDNN_PATH:$LD_LIBRARY_PATH
# only query the gpu used
nohup stdbuf -oL nvidia-smi \
--id=${CUDA_VISIBLE_DEVICES} \
--query-gpu=timestamp \
--query-compute-apps=pid,process_name,used_memory \
--format=csv \
--filename=mem.log \
-l 1 &
# mnist
# mnist gpu mnist 128
FLAGS_benchmark=true stdbuf -oL python fluid/mnist.py \
--device=GPU \
--batch_size=128 \
--skip_batch_num=5 \
--iterations=500 \
2>&1 | tee -a mnist_gpu_128.log
# vgg16
# gpu cifar10 128
FLAGS_benchmark=true stdbuf -oL python fluid/vgg16.py \
--device=GPU \
--batch_size=128 \
--skip_batch_num=5 \
--iterations=30 \
2>&1 | tee -a vgg16_gpu_128.log
# flowers gpu 128
FLAGS_benchmark=true stdbuf -oL python fluid/vgg16.py \
--device=GPU \
--batch_size=32 \
--data_set=flowers \
--skip_batch_num=5 \
--iterations=30 \
2>&1 | tee -a vgg16_gpu_flowers_32.log
# resnet50
# resnet50 gpu cifar10 128
FLAGS_benchmark=true stdbuf -oL python fluid/resnet50.py \
--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
# resnet50 gpu flowers 64
FLAGS_benchmark=true stdbuf -oL python fluid/resnet50.py \
--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
# lstm
# lstm gpu imdb 32 # tensorflow only support batch=32
FLAGS_benchmark=true stdbuf -oL python fluid/stacked_dynamic_lstm.py \
--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
# seq2seq
# seq2seq gpu wmb 128
FLAGS_benchmark=true stdbuf -oL python fluid/machine_translation.py \
--device=GPU \
--batch_size=128 \
--skip_batch_num=5 \
--iterations=30 \
2>&1 | tee -a lstm_gpu_128.log
benchmark/fluid/stacked_dynamic_lstm.py 0 → 100644
浏览文件 @ d2ba05a6
# 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.v2 as paddle
import paddle.v2.dataset.imdb as imdb
import paddle.fluid as fluid
from paddle.v2 import batch
import paddle.fluid.profiler as profiler
def parse_args():
parser = argparse.ArgumentParser("Understand Sentiment by Dynamic RNN.")
parser.add_argument(
'--batch_size',
type=int,
default=32,
help='The sequence number of a 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(
'--emb_dim',
type=int,
default=512,
help='Dimension of embedding table. (default: %(default)d)')
parser.add_argument(
'--hidden_dim',
type=int,
default=512,
help='Hidden size of lstm unit. (default: %(default)d)')
parser.add_argument(
'--pass_num',
type=int,
default=100,
help='Epoch number to train. (default: %(default)d)')
parser.add_argument(
'--device',
type=str,
default='CPU',
choices=['CPU', 'GPU'],
help='The device type.')
parser.add_argument(
'--crop_size',
type=int,
default=int(os.environ.get('CROP_SIZE', '1500')),
help='The max sentence length of input. Since this model use plain RNN,'
' Gradient could be explored if sentence is too long')
parser.add_argument(
'--with_test',
action='store_true',
help='If set, test the testset during training.')
args = parser.parse_args()
return args
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 main():
args = parse_args()
lstm_size = args.hidden_dim
data = fluid.layers.data(
name="words", shape=[1], lod_level=1, dtype='int64')
sentence = fluid.layers.embedding(
input=data, size=[len(word_dict), args.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()
adam.minimize(loss)
fluid.memory_optimize(fluid.default_main_program())
place = fluid.CPUPlace() if args.device == 'CPU' else fluid.CUDAPlace(0)
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
train_reader = batch(
paddle.reader.shuffle(
crop_sentence(imdb.train(word_dict), args.crop_size),
buf_size=25000),
batch_size=args.batch_size)
iters, num_samples, start_time = 0, 0, time.time()
for pass_id in range(args.pass_num):
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
tensor_words = to_lodtensor([x[0] for x in data], place)
label = numpy.array([x[1] for x in data]).astype("int64")
label = label.reshape((-1, 1))
loss_np, acc, weight = exe.run(
fluid.default_main_program(),
feed={"words": tensor_words,
"label": label},
fetch_list=[loss, batch_acc, batch_size_tensor])
iters += 1
for x in data:
num_samples += len(x[0])
print(
"Pass = %d, Iter = %d, Loss = %f, Accuracy = %f" %
(pass_id, iters, loss_np, acc)
) # 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))
exit(0)
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 = numpy.concatenate(data, axis=0).astype("int64")
flattened_data = flattened_data.reshape([len(flattened_data), 1])
res = fluid.LoDTensor()
res.set(flattened_data, place)
res.set_lod([lod])
return res
def print_arguments(args):
print('----------- lstm 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)
main()
benchmark/fluid/vgg.py 0 → 100644
浏览文件 @ d2ba05a6
# 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 argparse
import functools
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
'--batch_size', type=int, default=128, help="Batch size for training.")
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(
'--learning_rate',
type=float,
default=1e-3,
help="Learning rate for training.")
parser.add_argument('--pass_num', type=int, default=50, help="No. of passes.")
parser.add_argument(
'--device',
type=str,
default='GPU',
choices=['CPU', 'GPU'],
help="The device type.")
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(
'--with_test',
action='store_true',
help='If set, test the testset during training.')
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=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 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_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)
opts = optimizer.minimize(avg_cost)
fluid.memory_optimize(fluid.default_main_program())
# Initialize executor
place = core.CPUPlace() if args.device == 'CPU' else core.CUDAPlace(0)
exe = fluid.Executor(place)
# 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)
# test
def test(exe):
test_accuracy = 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])
acc, weight = exe.run(inference_program,
feed={"pixel": img_data,
"label": y_data},
fetch_list=[batch_acc, batch_size_tensor])
test_accuracy.add(value=acc, weight=weight)
return test_accuracy.eval()
iters, num_samples, start_time = 0, 0, time.time()
accuracy = fluid.average.WeightedAverage()
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(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, weight = exe.run(
fluid.default_main_program(),
feed={"pixel": img_data,
"label": y_data},
fetch_list=[avg_cost, batch_acc, batch_size_tensor])
accuracy.add(value=acc, weight=weight)
iters += 1
num_samples += len(y_data)
print(
"Pass = %d, Iter = %d, Loss = %f, Accuracy = %f" %
(pass_id, iters, loss, acc)
) # The accuracy is the accumulation of batches, but not the current batch.
# pass_train_acc = accuracy.eval()
train_losses.append(loss)
train_accs.append(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:
pass_test_acc = test(exe)
exit(0)
def print_arguments():
print('----------- vgg Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
if __name__ == "__main__":
print_arguments()
main()
benchmark/tensorflow/machine_translation.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
benchmark/tensorflow/mnist.py 0 → 100644
浏览文件 @ d2ba05a6
# 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 time
import numpy as np
import tensorflow as tf
import paddle.v2 as paddle
DTYPE = tf.float32
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(
'--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.')
args = parser.parse_args()
return args
def run_benchmark(args):
def weight_variable(dtype, shape):
initial = tf.truncated_normal(shape, stddev=0.1, dtype=dtype)
return tf.Variable(initial)
def bias_variable(dtype, shape):
initial = tf.constant(0.1, shape=shape, dtype=dtype)
return tf.Variable(initial)
device = '/cpu:0' if args.device == 'CPU' else '/device:GPU:0'
with tf.device(device):
images = tf.placeholder(DTYPE, shape=(None, 28, 28, 1))
labels = tf.placeholder(tf.int64, shape=(None, ))
# conv1, relu, pool1
conv1_weights = weight_variable(DTYPE, [5, 5, 1, 20])
conv1_bias = bias_variable(DTYPE, [20])
conv1 = tf.nn.conv2d(
images, conv1_weights, strides=[1, 1, 1, 1], padding="VALID")
relu1 = tf.nn.relu(tf.nn.bias_add(conv1, conv1_bias))
pool1 = tf.nn.max_pool(
relu1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="VALID")
# conv2, relu, pool2
conv2_weights = weight_variable(DTYPE, [5, 5, 20, 50])
conv2_bias = bias_variable(DTYPE, [50])
conv2 = tf.nn.conv2d(
pool1, conv2_weights, strides=[1, 1, 1, 1], padding="VALID")
relu2 = tf.nn.relu(tf.nn.bias_add(conv2, conv2_bias))
pool2 = tf.nn.max_pool(
relu2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="VALID")
# FC
pool_shape = pool2.get_shape().as_list()
hidden_dim = reduce(lambda a, b: a * b, pool_shape[1:], 1)
reshape = tf.reshape(pool2, shape=(tf.shape(pool2)[0], hidden_dim))
fc_weights = weight_variable(DTYPE, [hidden_dim, 10])
fc_bias = bias_variable(DTYPE, [10])
logits = tf.matmul(reshape, fc_weights) + fc_bias
# Get prediction
prediction = tf.nn.softmax(logits)
# Loss
one_hot_labels = tf.one_hot(labels, depth=10)
cost = -tf.reduce_sum(tf.log(prediction) * one_hot_labels, [1])
avg_cost = tf.reduce_mean(cost)
# Get accuracy
correct = tf.equal(tf.argmax(prediction, 1), labels)
accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))
# metrics, g_accuracy
with tf.variable_scope("reset_metrics_accuracy_scope") as scope:
g_accuracy = tf.metrics.accuracy(
labels, tf.argmax(
prediction, axis=1))
vars = tf.contrib.framework.get_variables(
scope, collection=tf.GraphKeys.LOCAL_VARIABLES)
g_accuracy_reset_op = tf.variables_initializer(vars)
# Optimizer
opt = tf.train.AdamOptimizer(
learning_rate=0.001, beta1=0.9, beta2=0.999)
train_op = opt.minimize(avg_cost)
# train_op = tf.train.AdamOptimizer(1e-4).minimize(avg_cost)
train_reader = paddle.batch(
paddle.dataset.mnist.train(), batch_size=args.batch_size)
test_reader = paddle.batch(
paddle.dataset.mnist.test(), batch_size=args.batch_size)
def eval_test():
sess.run(g_accuracy_reset_op)
for batch_id, data in enumerate(test_reader()):
images_data = np.array(
map(lambda x: np.transpose(x[0].reshape([1, 28, 28]), axes=[1,2,0]), data)).astype("float32")
labels_data = np.array(map(lambda x: x[1], data)).astype("int64")
loss, acc, g_acc = sess.run(
[avg_cost, accuracy, g_accuracy],
feed_dict={images: images_data,
labels: labels_data})
return g_acc[1]
config = tf.ConfigProto(
intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
init_g = tf.global_variables_initializer()
init_l = tf.local_variables_initializer()
sess.run(init_g)
sess.run(init_l)
for pass_id in range(args.pass_num):
sess.run(g_accuracy_reset_op)
pass_start = time.time()
for batch_id, data in enumerate(train_reader()):
images_data = np.array(
map(lambda x: np.transpose(x[0].reshape([1, 28, 28]), axes=[1,2,0]), data)).astype("float32")
labels_data = np.array(map(lambda x: x[1], data)).astype(
"int64")
start = time.time()
_, loss, acc, g_acc = sess.run(
[train_op, avg_cost, accuracy, g_accuracy],
feed_dict={images: images_data,
labels: labels_data})
end = time.time()
print("pass=%d, batch=%d, loss=%f, error=%f, elapse=%f" %
(pass_id, batch_id, loss, 1 - acc, (end - start) / 1000))
pass_end = time.time()
test_avg_acc = eval_test()
print(
"pass=%d, training_avg_accuracy=%f, test_avg_acc=%f, elapse=%f"
% (pass_id, g_acc[1], test_avg_acc,
(pass_end - pass_start) / 1000))
def print_arguments(args):
print('----------- 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)
run_benchmark(args)
benchmark/tensorflow/resnet.py 0 → 100644
浏览文件 @ d2ba05a6
# 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.
"""
based on https://github.com/tensorflow/models/blob/master/official/resnet/resnet_model.py
Get help: python resnet.py --help
See performance on flowers: python resnet.py
Train on cifar10: python resnet.py --data=cifar10 --with_test
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import time
import numpy as np
import paddle.v2 as paddle
import tensorflow as tf
DTYPE = tf.float32
def parse_args():
parser = argparse.ArgumentParser('Convolution model benchmark.')
parser.add_argument(
'--model',
type=str,
choices=['resnet'],
default='resnet',
help='The model architecture.')
parser.add_argument(
'--batch_size', type=int, default=32, help='The minibatch size.')
parser.add_argument(
'--use_fake_data',
action='store_true',
help='use real data or fake data')
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=105,
help='The number of minibatches.')
parser.add_argument(
'--pass_num', type=int, default=300, help='The number of passes.')
parser.add_argument(
'--order',
type=str,
default='NHWC',
choices=['NCHW', 'NHWC'],
help='The data order, now only support NCHW.')
parser.add_argument(
'--device',
type=str,
default='GPU',
choices=['CPU', 'GPU'],
help='The device type.')
parser.add_argument(
'--data',
type=str,
default='flowers102',
choices=['flowers102', 'cifar10'],
help='The kinds of data.')
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(
'--with_test',
action='store_true',
help='If set, test the testset during training.')
parser.add_argument(
'--use_nvprof',
action='store_true',
help='If set, use nvprof for CUDA.')
args = parser.parse_args()
return args
def print_arguments(args):
vars(args)['use_nvprof'] = (vars(args)['use_nvprof'] and
vars(args)['device'] == 'GPU')
vars(args)['iterations'] = vars(args)['pass_num'] * 1000 if vars(args)[
'with_test'] else vars(args)['iterations']
print('----------- Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
def fixed_padding(inputs, kernel_size, data_format):
"""Pads the input along the spatial dimensions independently of input size.
Args:
inputs: A tensor of size [batch, channels, height_in, width_in] or
[batch, height_in, width_in, channels] depending on data_format.
kernel_size: The kernel to be used in the conv2d or max_pool2d operation.
Should be a positive integer.
data_format: The input format ('channels_last' or 'channels_first').
Returns:
A tensor with the same format as the input with the data either intact
(if kernel_size == 1) or padded (if kernel_size > 1).
"""
pad_total = kernel_size - 1
pad_beg = pad_total // 2
pad_end = pad_total - pad_beg
if data_format == 'channels_first':
padded_inputs = tf.pad(inputs, [[0, 0], [0, 0], [pad_beg, pad_end],
[pad_beg, pad_end]])
else:
padded_inputs = tf.pad(inputs, [[0, 0], [pad_beg, pad_end],
[pad_beg, pad_end], [0, 0]])
return padded_inputs
def conv2d_fixed_padding(inputs, filters, kernel_size, strides, data_format):
"""Strided 2-D convolution with explicit padding."""
# The padding is consistent and is based only on `kernel_size`, not on the
# dimensions of `inputs` (as opposed to using `tf.layers.conv2d` alone).
# This is consistent with PaddlePaddle.
# In addition, the calculation for output size in TensorFlow can refer:
# https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/framework/common_shape_fns.cc
if strides > 1:
inputs = fixed_padding(inputs, kernel_size, data_format)
return tf.layers.conv2d(
inputs=inputs,
filters=filters,
kernel_size=kernel_size,
strides=strides,
padding=('SAME' if strides == 1 else 'VALID'),
use_bias=False,
kernel_initializer=tf.variance_scaling_initializer(),
data_format=data_format)
def conv_bn(inputs,
filters,
kernel_size,
strides,
is_training,
data_format,
act=True):
# def conv2d_fixed_padding(inputs, filters, kernel_size, strides, data_format):
# set fused=True for a significant performance boost. See
# https://www.tensorflow.org/performance/performance_guide#common_fused_ops
inputs = conv2d_fixed_padding(
inputs=inputs,
filters=filters,
kernel_size=kernel_size,
strides=strides,
data_format=data_format)
inputs = tf.layers.batch_normalization(
inputs=inputs,
axis=1 if data_format == 'channels_first' else 3,
momentum=0.9,
epsilon=1e-05,
center=True,
scale=True,
training=is_training,
fused=True)
if act:
inputs = tf.nn.relu(inputs)
return inputs
def basicblock(inputs, filters, is_training, projection_shortcut, strides,
data_format):
shortcut = inputs
if projection_shortcut is not None:
shortcut = projection_shortcut(inputs)
inputs = conv_bn(inputs, filters, 3, strides, is_training, data_format)
inputs = conv_bn(inputs, filters, 3, 1, is_training, data_format, act=False)
inputs = inputs + shortcut
inputs = tf.nn.relu(inputs)
return inputs
def bottleneck(inputs, filters, is_training, projection_shortcut, strides,
data_format):
shortcut = inputs
if projection_shortcut is not None:
shortcut = projection_shortcut(inputs)
inputs = conv_bn(inputs, filters, 1, strides, is_training, data_format)
inputs = conv_bn(inputs, filters, 3, 1, is_training, data_format, act=False)
inputs = conv_bn(
inputs, filters * 4, 1, 1, is_training, data_format, act=False)
inputs = inputs + shortcut
inputs = tf.nn.relu(inputs)
return inputs
def block_layer(inputs, filters, block_fn, blocks, strides, is_training, name,
data_format):
# Bottleneck blocks end with 4x the number of filters as they start with
filters_out = 4 * filters if block_fn is bottleneck else filters
def projection_shortcut(inputs):
return conv2d_fixed_padding(
inputs=inputs,
filters=filters_out,
kernel_size=1,
strides=strides,
data_format=data_format)
# Only the first block per block_layer uses projection_shortcut and strides
inputs = block_fn(inputs, filters, is_training, projection_shortcut,
strides, data_format)
for _ in range(1, blocks):
inputs = block_fn(inputs, filters, is_training, None, 1, data_format)
return tf.identity(inputs, name)
def resnet_imagenet(depth, class_dim, data_format):
"""Returns the ResNet model for a given size and number of output classes."""
def resnet_generator(block_fn,
layers,
num_classes,
data_format='channels_last'):
if data_format is None:
data_format = ('channels_first'
if tf.test.is_built_with_cuda() else 'channels_last')
def model(inputs, is_training):
"""Constructs the ResNet model given the inputs."""
if data_format == 'channels_first':
# Convert the inputs from channels_last (NHWC) to channels_first (NCHW).
# This provides a large performance boost on GPU. See
# https://www.tensorflow.org/performance/performance_guide#data_formats
inputs = tf.transpose(inputs, [0, 3, 1, 2])
inputs = conv_bn(inputs, 64, 7, 2, is_training, data_format)
inputs = tf.identity(inputs, 'initial_conv')
inputs = tf.layers.max_pooling2d(
inputs=inputs,
pool_size=3,
strides=2,
padding='SAME',
data_format=data_format)
inputs = tf.identity(inputs, 'initial_max_pool')
inputs = block_layer(inputs, 64, block_fn, layers[0], 1,
is_training, 'block_layer1', data_format)
inputs = block_layer(inputs, 128, block_fn, layers[1], 2,
is_training, 'block_layer2', data_format)
inputs = block_layer(inputs, 256, block_fn, layers[2], 2,
is_training, 'block_layer3', data_format)
inputs = block_layer(inputs, 512, block_fn, layers[3], 2,
is_training, 'block_layer4', data_format)
inputs = tf.layers.average_pooling2d(
inputs=inputs,
pool_size=7,
strides=1,
padding='VALID',
data_format=data_format)
inputs = tf.identity(inputs, 'final_avg_pool')
inputs = tf.reshape(inputs,
[-1, 512 if block_fn is basicblock else 2048])
inputs = tf.layers.dense(inputs=inputs, units=num_classes)
inputs = tf.identity(inputs, 'final_dense')
return inputs
return model
model_params = {
18: {
'block': basicblock,
'layers': [2, 2, 2, 2]
},
34: {
'block': basicblock,
'layers': [3, 4, 6, 3]
},
50: {
'block': bottleneck,
'layers': [3, 4, 6, 3]
},
101: {
'block': bottleneck,
'layers': [3, 4, 23, 3]
},
152: {
'block': bottleneck,
'layers': [3, 8, 36, 3]
},
200: {
'block': bottleneck,
'layers': [3, 24, 36, 3]
}
}
if depth not in model_params:
raise ValueError('Not a valid depth:', depth)
params = model_params[depth]
return resnet_generator(params['block'], params['layers'], class_dim,
data_format)
def resnet_cifar10(depth, num_classes, data_format):
if depth % 6 != 2:
raise ValueError('depth must be 6n + 2:', depth)
num_blocks = (depth - 2) // 6
if data_format is None:
data_format = ('channels_first'
if tf.test.is_built_with_cuda() else 'channels_last')
def model(inputs, is_training):
inputs = conv_bn(inputs, 16, 3, 1, is_training, data_format)
inputs = tf.identity(inputs, 'initial_conv')
inputs = block_layer(inputs, 16, basicblock, num_blocks, 1, is_training,
'block_layer1', data_format)
inputs = block_layer(inputs, 32, basicblock, num_blocks, 2, is_training,
'block_layer2', data_format)
inputs = block_layer(inputs, 64, basicblock, num_blocks, 2, is_training,
'block_layer3', data_format)
inputs = tf.layers.average_pooling2d(
inputs=inputs,
pool_size=8,
strides=1,
padding='VALID',
data_format=data_format)
inputs = tf.identity(inputs, 'final_avg_pool')
inputs = tf.reshape(inputs, [-1, 64])
inputs = tf.layers.dense(inputs=inputs, units=num_classes)
inputs = tf.identity(inputs, 'final_dense')
return inputs
return model
def run_benchmark(args, data_format='channels_last', device='/cpu:0'):
"""Our model_fn for ResNet to be used with our Estimator."""
class_dim = 1000
dshape = (None, 224, 224, 3)
pdshape = (3, 224, 224)
if args.data == 'flowers102':
class_dim = 102
dshape = (None, 224, 224, 3)
pdshape = (3, 224, 224)
elif args.data == 'cifar10':
class_dim = 10
dshape = (None, 32, 32, 3)
pdshape = (3, 32, 32)
with tf.device(device):
images = tf.placeholder(DTYPE, shape=dshape)
labels = tf.placeholder(tf.int64, shape=(None, ))
is_training = tf.placeholder('bool')
onehot_labels = tf.one_hot(labels, depth=class_dim)
network = resnet_cifar10(
32, class_dim,
data_format) if args.data == 'cifar10' else resnet_imagenet(
50, class_dim, data_format)
logits = network(inputs=images, is_training=is_training)
cross_entropy = tf.losses.softmax_cross_entropy(
logits=logits, onehot_labels=onehot_labels)
avg_cost = tf.reduce_mean(cross_entropy)
correct = tf.equal(tf.argmax(logits, 1), labels)
accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))
lr = 0.1 if args.data == 'cifar10' else 0.01
optimizer = tf.train.MomentumOptimizer(learning_rate=lr, momentum=0.9)
# Batch norm requires update_ops to be added as a train_op dependency.
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
train_op = optimizer.minimize(avg_cost)
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.cifar.train10()
if args.data == '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 == 'cifar10' else paddle.dataset.flowers.test(),
batch_size=100)
def test():
test_accs = []
for batch_id, data in enumerate(test_reader()):
test_images = np.array(
map(lambda x: np.transpose(x[0].reshape(pdshape),
axes=[1, 2, 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
}))
print("Pass = %d, Train performance = %f imgs/s, Test accuracy = %f\n" %
(pass_id, num_samples / train_elapsed, np.mean(test_accs)))
config = tf.ConfigProto(
intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
init_g = tf.global_variables_initializer()
init_l = tf.local_variables_initializer()
sess.run(init_g)
sess.run(init_l)
if args.use_fake_data:
data = train_reader().next()
images_data = np.array(
map(lambda x: np.transpose(x[0].reshape(pdshape),
axes=[1, 2, 0]), data)).astype("float32")
labels_data = np.array(map(lambda x: x[1], data)).astype('int64')
iters, num_samples, start_time = 0, 0, 0.0
for pass_id in range(args.pass_num):
if iters == args.iterations:
break
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
if not args.use_fake_data:
images_data = np.array(
map(lambda x: np.transpose(x[0].reshape(pdshape),
axes=[1, 2, 0]), data)).astype("float32")
labels_data = np.array(map(lambda x: x[1], data)).astype(
'int64')
_, loss, acc = sess.run([train_op, avg_cost, accuracy],
feed_dict={
images: images_data,
labels: labels_data,
is_training: True
})
iters += 1
train_accs.append(acc)
train_losses.append(loss)
num_samples += len(data)
print("Pass=%d, Iter=%d, Loss=%f, Accuray=%f\n" %
(pass_id, iters, loss, acc))
train_elapsed = time.time() - start_time
print("Pass=%d, Loss=%f, Accuray=%f\n" %
(pass_id, np.mean(train_losses), np.mean(train_accs)))
# evaluation
if args.with_test:
test()
if not args.with_test:
duration = time.time() - start_time
examples_per_sec = num_samples / duration
sec_per_batch = duration / (iters - args.skip_batch_num)
print('Total examples: %d, total time: %.5f' %
(num_samples, duration))
print('%.5f examples/sec, %.5f sec/batch' %
(examples_per_sec, sec_per_batch))
if __name__ == '__main__':
args = parse_args()
print_arguments(args)
if tf.test.is_built_with_cuda():
device = '/device:GPU:0'
if args.order == 'NHWC':
data_format = 'channels_last'
else:
data_format = 'channels_first'
else:
device = '/cpu:0'
if args.order == 'NHWC':
data_format = 'channels_last'
else:
raise ValueError('Only support NHWC order in CPU mode')
run_benchmark(args, data_format, device)
benchmark/tensorflow/stacked_dynamic_lstm.py 0 → 100644
浏览文件 @ d2ba05a6
# 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 tensorflow as tf
import paddle.v2 as paddle
def parse_args():
parser = argparse.ArgumentParser("LSTM model benchmark.")
parser.add_argument(
'--batch_size',
type=int,
default=32,
help='The sequence number of a batch data. (default: %(default)d)')
parser.add_argument(
'--stacked_num',
type=int,
default=5,
help='Number of lstm layers to stack. (default: %(default)d)')
parser.add_argument(
'--embedding_dim',
type=int,
default=512,
help='Dimension of embedding table. (default: %(default)d)')
parser.add_argument(
'--hidden_dim',
type=int,
default=512,
help='Hidden size of lstm unit. (default: %(default)d)')
parser.add_argument(
'--pass_num',
type=int,
default=10,
help='Epoch number to train. (default: %(default)d)')
parser.add_argument(
'--learning_rate',
type=float,
default=0.0002,
help='Learning rate used to train. (default: %(default)f)')
parser.add_argument(
'--infer_only', action='store_true', help='If set, run forward only.')
args = parser.parse_args()
return args
def print_arguments(args):
print('----------- Configuration Arguments -----------')
for arg, value in sorted(vars(args).iteritems()):
print('%s: %s' % (arg, value))
print('------------------------------------------------')
def dynamic_lstm_model(dict_size,
embedding_dim,
hidden_dim,
stacked_num,
class_num=2,
is_train=True):
word_idx = tf.placeholder(tf.int64, shape=[None, None])
sequence_length = tf.placeholder(tf.int64, shape=[None, ])
embedding_weights = tf.get_variable('word_embeddings',
[dict_size, embedding_dim])
embedding = tf.nn.embedding_lookup(embedding_weights, word_idx)
lstm_cell = tf.nn.rnn_cell.LSTMCell(
num_units=hidden_dim, use_peepholes=False)
stacked_cell = tf.nn.rnn_cell.MultiRNNCell([lstm_cell] * stacked_num)
# final_state [LSTMTuple(c, h), LSTMTuple(c, h) ...] total stacked_num LSTMTuples
_, final_state = tf.nn.dynamic_rnn(
cell=stacked_cell,
inputs=embedding,
dtype=tf.float32,
sequence_length=sequence_length)
w = tf.Variable(
tf.truncated_normal([hidden_dim, class_num]), dtype=tf.float32)
bias = tf.Variable(
tf.constant(
value=0.0, shape=[class_num], dtype=tf.float32))
prediction = tf.matmul(final_state[-1][1], w) + bias
if not is_train:
return (word_idx, sequence_length), tf.nn.softmax(prediction)
label = tf.placeholder(tf.int64, shape=[None, ])
loss = tf.nn.softmax_cross_entropy_with_logits(
labels=tf.one_hot(label, 2), logits=prediction)
avg_loss = tf.reduce_mean(loss)
correct_count = tf.equal(tf.argmax(prediction, 1), label)
acc = tf.reduce_mean(tf.cast(correct_count, tf.float32))
with tf.variable_scope("reset_metrics_accuracy_scope") as scope:
g_acc = tf.metrics.accuracy(label, tf.argmax(prediction, axis=1))
vars = tf.contrib.framework.get_variables(
scope, collection=tf.GraphKeys.LOCAL_VARIABLES)
reset_op = tf.variables_initializer(vars)
return (word_idx, sequence_length, label), avg_loss, acc, g_acc, reset_op
def padding_data(data, padding_size, value):
data = data + [value] * padding_size
return data[:padding_size]
def train(args):
word_dict = paddle.dataset.imdb.word_dict()
dict_size = len(word_dict)
feeding_list, avg_loss, acc, g_acc, reset_op = dynamic_lstm_model(
dict_size, args.embedding_dim, args.hidden_dim, args.stacked_num)
adam_optimizer = tf.train.AdamOptimizer(learning_rate=args.learning_rate)
train_op = adam_optimizer.minimize(avg_loss)
train_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.imdb.train(word_dict), buf_size=25000),
batch_size=args.batch_size)
test_reader = paddle.batch(
paddle.reader.shuffle(
paddle.dataset.imdb.test(word_dict), buf_size=25000),
batch_size=args.batch_size)
def do_validation(sess):
sess.run(reset_op)
for batch_id, data in enumerate(test_reader()):
word_idx = map(lambda x: x[0], data)
sequence_length = np.array(
[len(seq) for seq in word_idx]).astype('int64')
maxlen = np.max(sequence_length)
word_idx = [padding_data(seq, maxlen, 0) for seq in word_idx]
word_idx = np.array(word_idx).astype('int64')
label = np.array(map(lambda x: x[1], data)).astype('int64')
_, loss, fetch_acc, fetch_g_acc = sess.run(
[train_op, avg_loss, acc, g_acc],
feed_dict={
feeding_list[0]: word_idx,
feeding_list[1]: sequence_length,
feeding_list[2]: label
})
return fetch_g_acc[1]
config = tf.ConfigProto(
intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
init_g = tf.global_variables_initializer()
init_l = tf.local_variables_initializer()
sess.run(init_l)
sess.run(init_g)
for pass_id in xrange(args.pass_num):
# clear accuracy local variable
sess.run(reset_op)
pass_start_time = time.time()
words_seen = 0
for batch_id, data in enumerate(train_reader()):
word_idx = map(lambda x: x[0], data)
sequence_length = np.array(
[len(seq) for seq in word_idx]).astype('int64')
words_seen += np.sum(sequence_length)
maxlen = np.max(sequence_length)
word_idx = [padding_data(seq, maxlen, 0) for seq in word_idx]
word_idx = np.array(word_idx).astype('int64')
label = np.array(map(lambda x: x[1], data)).astype('int64')
_, loss, fetch_acc, fetch_g_acc = sess.run(
[train_op, avg_loss, acc, g_acc],
feed_dict={
feeding_list[0]: word_idx,
feeding_list[1]: sequence_length,
feeding_list[2]: label
})
print("pass_id=%d, batch_id=%d, loss: %f, acc: %f, avg_acc: %f"
% (pass_id, batch_id, loss, fetch_acc, fetch_g_acc[1]))
pass_end_time = time.time()
time_consumed = pass_end_time - pass_start_time
words_per_sec = words_seen / time_consumed
test_acc = do_validation(sess)
print("pass_id=%d, test_acc: %f, words/s: %f, sec/pass: %f" %
(pass_id, test_acc, words_per_sec, time_consumed))
if __name__ == '__main__':
args = parse_args()
print_arguments(args)
if args.infer_only:
pass
else:
train(args)
benchmark/tensorflow/vgg.py 0 → 100644
浏览文件 @ d2ba05a6
# 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"""
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(
'--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(
'--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='GPU',
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.')
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():
"""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 = '/cpu:0' if args.device == 'CPU' else '/device:GPU:0'
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)
with tf.control_dependencies(update_ops):
train_op = optimizer.minimize(avg_loss)
# 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)
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
init_g = tf.global_variables_initializer()
init_l = tf.local_variables_initializer()
sess.run(init_g)
sess.run(init_l)
iters, num_samples, start_time = 0, 0, time.time()
for pass_id in range(args.num_passes):
# train
num_samples = 0
start_time = time.time()
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
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')
_, loss, acc = sess.run([train_op, avg_loss, accuracy],
feed_dict={
images: train_images,
labels: train_labels,
is_training: True
})
iters += 1
num_samples += len(data)
print("Pass = %d, Iters = %d, Loss = %f, Accuracy = %f" %
(pass_id, iters, loss, acc))
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()
run_benchmark()
cmake/cblas.cmake
浏览文件 @ d2ba05a6
......@@ -62,29 +62,33 @@ endif()
## Then find the reference-cblas. www.netlib.org/blas/
set(REFERENCE_CBLAS_ROOT $ENV{REFERENCE_CBLAS_ROOT} CACHE PATH
"Folder contains reference-cblas")
set(REFERENCE_CBLAS_INCLUDE_SEARCH_PATHS
${REFERENCE_CBLAS_ROOT}/include
/usr/include
/usr/include/cblas
)
set(REFERENCE_CBLAS_LIB_SEARCH_PATHS
${REFERENCE_CBLAS_ROOT}/lib
/usr/lib
/usr/lib/blas/reference/
/usr/lib/reference/
)
if(NOT CMAKE_CROSSCOMPILING)
set(REFERENCE_CBLAS_INCLUDE_SEARCH_PATHS
${REFERENCE_CBLAS_ROOT}/include
/usr/include
/usr/include/cblas
)
set(REFERENCE_CBLAS_LIB_SEARCH_PATHS
${REFERENCE_CBLAS_ROOT}/lib
/usr/lib
/usr/lib/blas/reference/
/usr/lib/reference/
)
else()
# Disable the finding of reference cblas under host's system path
set(REFERENCE_CBLAS_INCLUDE_SEARCH_PATHS ${REFERENCE_CBLAS_ROOT}/include)
set(REFERENCE_CBLAS_LIB_SEARCH_PATHS ${REFERENCE_CBLAS_ROOT}/lib)
endif()
find_path(REFERENCE_CBLAS_INCLUDE_DIR NAMES cblas.h PATHS
${REFERENCE_CBLAS_INCLUDE_SEARCH_PATHS})
find_library(REFERENCE_CBLAS_LIBRARY NAMES cblas PATHS
${REFERENCE_CBLAS_LIB_SEARCH_PATHS})
if (REFERENCE_CBLAS_INCLUDE_DIR AND REFERENCE_CBLAS_LIBRARY)
if(REFERENCE_CBLAS_INCLUDE_DIR AND REFERENCE_CBLAS_LIBRARY)
set(CBLAS_FOUND ON)
set(CBLAS_PROVIDER REFERENCE)
set(CBLAS_INC_DIR ${REFERENCE_CBLAS_INCLUDE_DIR})
......
cmake/external/grpc.cmake
浏览文件 @ d2ba05a6
......@@ -24,16 +24,16 @@ SET(GRPC_INSTALL_DIR ${THIRD_PARTY_PATH}/install/grpc)
SET(GRPC_INCLUDE_DIR "${GRPC_INSTALL_DIR}/include/" CACHE PATH "grpc include directory." FORCE)
SET(GRPC_CPP_PLUGIN "${GRPC_INSTALL_DIR}/bin/grpc_cpp_plugin" CACHE FILEPATH "GRPC_CPP_PLUGIN" FORCE)
IF(APPLE)
SET(BUILD_CMD make -n HAS_SYSTEM_PROTOBUF=false -s -j8 static grpc_cpp_plugin | sed "s/-Werror//g" | sh)
SET(BUILD_CMD make -n HAS_SYSTEM_PROTOBUF=false -s -j static grpc_cpp_plugin | sed "s/-Werror//g" | sh)
ELSE()
SET(BUILD_CMD make HAS_SYSTEM_PROTOBUF=false -s -j8 static grpc_cpp_plugin)
SET(BUILD_CMD make HAS_SYSTEM_PROTOBUF=false -s -j static grpc_cpp_plugin)
ENDIF()
ExternalProject_Add(
extern_grpc
DEPENDS protobuf zlib
GIT_REPOSITORY "https://github.com/grpc/grpc.git"
GIT_TAG "v1.8.x"
GIT_TAG "v1.10.x"
PREFIX ${GRPC_SOURCES_DIR}
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
......
cmake/external/mkldnn.cmake
浏览文件 @ d2ba05a6
......@@ -36,7 +36,8 @@ MESSAGE(STATUS "Set ${MKLDNN_INSTALL_DIR}/lib to runtime path")
SET(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE)
SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${MKLDNN_INSTALL_DIR}/lib")
INCLUDE_DIRECTORIES(${MKLDNN_INC_DIR})
INCLUDE_DIRECTORIES(${MKLDNN_INC_DIR}) # For MKLDNN code to include internal headers.
INCLUDE_DIRECTORIES(${THIRD_PARTY_PATH}/install) # For Paddle code to include mkldnn.h
IF(${CBLAS_PROVIDER} STREQUAL "MKLML")
SET(MKLDNN_DEPENDS ${MKLML_PROJECT})
......
cmake/external/mklml.cmake
浏览文件 @ d2ba05a6
......@@ -28,7 +28,7 @@ INCLUDE(ExternalProject)
SET(MKLML_PROJECT "extern_mklml")
SET(MKLML_VER "mklml_lnx_2018.0.1.20171007")
SET(MKLML_URL "https://github.com/01org/mkl-dnn/releases/download/v0.11/${MKLML_VER}.tgz")
SET(MKLML_URL "http://paddlepaddledeps.bj.bcebos.com/${MKLML_VER}.tgz")
SET(MKLML_SOURCE_DIR "${THIRD_PARTY_PATH}/mklml")
SET(MKLML_DOWNLOAD_DIR "${MKLML_SOURCE_DIR}/src/${MKLML_PROJECT}")
SET(MKLML_DST_DIR "mklml")
......
cmake/external/nccl.cmake 已删除 100644 → 0
浏览文件 @ 10669f1f
# 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.
if(NOT WITH_GPU)
return()
endif()
include(ExternalProject)
set(NCCL_SOURCE_DIR ${THIRD_PARTY_PATH}/nccl)
include_directories(${NCCL_SOURCE_DIR}/src/extern_nccl/src)
if(WITH_DSO)
# If we use DSO, we do not build nccl, just download the dependencies
set(NCCL_BUILD_COMMAND "")
set(NCCL_INSTALL_COMMAND "")
set(NCCL_INSTALL_DIR "")
else()
# otherwise, we build nccl and link it.
set(NCCL_INSTALL_DIR ${THIRD_PARTY_PATH}/install/nccl)
# Note: cuda 8.0 is needed to make nccl
# When cuda is not installed on the system directory, need to set CUDA_HOME to your cuda root
set(NCCL_BUILD_COMMAND "make -j 8")
set(NCCL_INSTALL_COMMAND "make install PREFIX=${NCCL_INSTALL_DIR}")
endif()
ExternalProject_Add(
extern_nccl
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/NVIDIA/nccl.git"
GIT_TAG "v1.3.4-1"
PREFIX "${NCCL_SOURCE_DIR}"
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_COMMAND "${NCCL_BUILD_COMMAND}"
INSTALL_COMMAND "${NCCL_INSTALL_COMMAND}"
INSTALL_DIR "${NCCL_INSTALL_DIR}"
TEST_COMMAND ""
)
if(WITH_DSO)
if(${CMAKE_VERSION} VERSION_LESS "3.3.0")
set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/lib_nccl_dummy.c)
file(WRITE ${dummyfile} "const char * dummy_nccl = \"${dummyfile}\";")
add_library(nccl STATIC ${dummyfile})
else()
add_library(nccl INTERFACE)
endif()
else()
add_library(nccl STATIC IMPORTED GLOBAL)
set_property(TARGET nccl PROPERTY IMPORTED_LOCATION
${NCCL_INSTALL_DIR}/lib/libnccl_static.a)
endif()
add_dependencies(nccl extern_nccl)
cmake/external/snappy.cmake
浏览文件 @ d2ba05a6
......@@ -11,19 +11,20 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
IF(MOBILE_INFERENCE)
if(MOBILE_INFERENCE OR RPI)
return()
ENDIF()
endif()
include (ExternalProject)
# NOTE: snappy is needed when linking with recordio
SET(SNAPPY_SOURCES_DIR ${THIRD_PARTY_PATH}/snappy)
SET(SNAPPY_INSTALL_DIR ${THIRD_PARTY_PATH}/install/snappy)
SET(SNAPPY_INCLUDE_DIR "${SNAPPY_INSTALL_DIR}/include/" CACHE PATH "snappy include directory." FORCE)
set(SNAPPY_SOURCES_DIR ${THIRD_PARTY_PATH}/snappy)
set(SNAPPY_INSTALL_DIR ${THIRD_PARTY_PATH}/install/snappy)
set(SNAPPY_INCLUDE_DIR "${SNAPPY_INSTALL_DIR}/include" CACHE PATH "snappy include directory." FORCE)
set(SNAPPY_LIBRARIES "${SNAPPY_INSTALL_DIR}/lib/libsnappy.a")
ExternalProject_Add(
extern_snappy
......@@ -51,8 +52,7 @@ ExternalProject_Add(
)
add_library(snappy STATIC IMPORTED GLOBAL)
set_property(TARGET snappy PROPERTY IMPORTED_LOCATION
"${SNAPPY_INSTALL_DIR}/lib/libsnappy.a")
set_property(TARGET snappy PROPERTY IMPORTED_LOCATION ${SNAPPY_LIBRARIES})
include_directories(${SNAPPY_INCLUDE_DIR})
add_dependencies(snappy extern_snappy)
cmake/external/snappystream.cmake
浏览文件 @ d2ba05a6
......@@ -11,9 +11,8 @@
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
IF(MOBILE_INFERENCE)
IF(MOBILE_INFERENCE OR RPI)
return()
ENDIF()
......@@ -21,9 +20,11 @@ include (ExternalProject)
# NOTE: snappy is needed when linking with recordio
SET(SNAPPYSTREAM_SOURCES_DIR ${THIRD_PARTY_PATH}/snappy_stream)
SET(SNAPPYSTREAM_INSTALL_DIR ${THIRD_PARTY_PATH}/install/snappy_stream)
SET(SNAPPYSTREAM_INCLUDE_DIR "${SNAPPYSTREAM_INSTALL_DIR}/include/" CACHE PATH "snappy stream include directory." FORCE)
set(SNAPPYSTREAM_SOURCES_DIR ${THIRD_PARTY_PATH}/snappy_stream)
set(SNAPPYSTREAM_INSTALL_DIR ${THIRD_PARTY_PATH}/install/snappy_stream)
set(SNAPPYSTREAM_INCLUDE_DIR "${SNAPPYSTREAM_INSTALL_DIR}/include" CACHE PATH "snappy stream include directory." FORCE)
set(SNAPPYSTREAM_LIBRARIES "${SNAPPYSTREAM_INSTALL_DIR}/lib/libsnappystream.a")
ExternalProject_Add(
extern_snappystream
......@@ -51,8 +52,9 @@ ExternalProject_Add(
)
add_library(snappystream STATIC IMPORTED GLOBAL)
set_property(TARGET snappystream PROPERTY IMPORTED_LOCATION
"${SNAPPYSTREAM_INSTALL_DIR}/lib/libsnappystream.a")
set_property(TARGET snappystream PROPERTY IMPORTED_LOCATION ${SNAPPYSTREAM_LIBRARIES})
include_directories(${SNAPPYSTREAM_INCLUDE_DIR}) # For snappysteam to include its own headers.
include_directories(${THIRD_PARTY_PATH}/install) # For Paddle to include snappy stream headers.
include_directories(${SNAPPYSTREAM_INCLUDE_DIR})
add_dependencies(snappystream extern_snappystream)
cmake/external/threadpool.cmake 0 → 100644
浏览文件 @ d2ba05a6
INCLUDE(ExternalProject)
SET(THREADPOOL_SOURCE_DIR ${THIRD_PARTY_PATH}/threadpool)
SET(THREADPOOL_INCLUDE_DIR ${THREADPOOL_SOURCE_DIR}/src/extern_threadpool)
INCLUDE_DIRECTORIES(${THREADPOOL_INCLUDE_DIR})
ExternalProject_Add(
extern_threadpool
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/progschj/ThreadPool.git"
GIT_TAG 9a42ec1329f259a5f4881a291db1dcb8f2ad9040
PREFIX ${THREADPOOL_SOURCE_DIR}
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND ""
TEST_COMMAND ""
)
if (${CMAKE_VERSION} VERSION_LESS "3.3.0")
set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/threadpool_dummy.c)
file(WRITE ${dummyfile} "const char *dummy_threadpool = \"${dummyfile}\";")
add_library(simple_threadpool STATIC ${dummyfile})
else()
add_library(simple_threadpool INTERFACE)
endif()
add_dependencies(simple_threadpool extern_threadpool)
LIST(APPEND external_project_dependencies simple_threadpool)
cmake/external/warpctc.cmake
浏览文件 @ d2ba05a6
......@@ -62,7 +62,8 @@ ExternalProject_Add(
)
MESSAGE(STATUS "warp-ctc library: ${WARPCTC_LIBRARIES}")
INCLUDE_DIRECTORIES(${WARPCTC_INCLUDE_DIR})
INCLUDE_DIRECTORIES(${WARPCTC_INCLUDE_DIR}) # For warpctc code to include its headers.
INCLUDE_DIRECTORIES(${THIRD_PARTY_PATH}/install) # For Paddle code to include warpctc headers.
ADD_LIBRARY(warpctc SHARED IMPORTED GLOBAL)
SET_PROPERTY(TARGET warpctc PROPERTY IMPORTED_LOCATION ${WARPCTC_LIBRARIES})
......
cmake/external/zlib.cmake
浏览文件 @ d2ba05a6
......@@ -25,7 +25,8 @@ ELSE(WIN32)
SET(ZLIB_LIBRARIES "${ZLIB_INSTALL_DIR}/lib/libz.a" CACHE FILEPATH "zlib library." FORCE)
ENDIF(WIN32)
INCLUDE_DIRECTORIES(${ZLIB_INCLUDE_DIR})
INCLUDE_DIRECTORIES(${ZLIB_INCLUDE_DIR}) # For zlib code to include its own headers.
INCLUDE_DIRECTORIES(${THIRD_PARTY_PATH}/install) # For Paddle code to include zlib.h.
ExternalProject_Add(
extern_zlib
......
cmake/generic.cmake
浏览文件 @ d2ba05a6
......@@ -195,14 +195,7 @@ function(cc_library TARGET_NAME)
list(REMOVE_ITEM cc_library_DEPS warpctc)
add_dependencies(${TARGET_NAME} warpctc)
endif()
if("${cc_library_DEPS}" MATCHES "ARCHIVE_START")
# Support linking flags: --whole-archive (Linux) / -force_load (MacOS).
# WARNING: Please don't use ARCHIVE_START&ARCHIVE_END if TARGET_NAME will be linked by other libraries.
target_circle_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
list(REMOVE_ITEM cc_library_DEPS ARCHIVE_START ARCHIVE_END)
else()
target_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
endif()
target_link_libraries(${TARGET_NAME} ${cc_library_DEPS})
add_dependencies(${TARGET_NAME} ${cc_library_DEPS})
endif()
......@@ -243,15 +236,11 @@ function(cc_test TARGET_NAME)
set(multiValueArgs SRCS DEPS ARGS)
cmake_parse_arguments(cc_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
add_executable(${TARGET_NAME} ${cc_test_SRCS})
# Support linking flags: --whole-archive (Linux) / -force_load (MacOS)
target_circle_link_libraries(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main paddle_memory gtest gflags glog)
if("${cc_test_DEPS}" MATCHES "ARCHIVE_START")
list(REMOVE_ITEM cc_test_DEPS ARCHIVE_START ARCHIVE_END)
endif()
add_dependencies(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main paddle_memory gtest gflags glog)
target_link_libraries(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main memory gtest gflags glog)
add_dependencies(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main memory gtest gflags glog)
add_test(NAME ${TARGET_NAME}
COMMAND ${TARGET_NAME} ${cc_test_ARGS}
WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
endif()
endfunction(cc_test)
......@@ -311,8 +300,8 @@ function(nv_test TARGET_NAME)
set(multiValueArgs SRCS DEPS)
cmake_parse_arguments(nv_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
cuda_add_executable(${TARGET_NAME} ${nv_test_SRCS})
target_link_libraries(${TARGET_NAME} ${nv_test_DEPS} paddle_gtest_main paddle_memory gtest gflags glog)
add_dependencies(${TARGET_NAME} ${nv_test_DEPS} paddle_gtest_main paddle_memory gtest gflags glog)
target_link_libraries(${TARGET_NAME} ${nv_test_DEPS} paddle_gtest_main memory gtest gflags glog)
add_dependencies(${TARGET_NAME} ${nv_test_DEPS} paddle_gtest_main memory gtest gflags glog)
add_test(${TARGET_NAME} ${TARGET_NAME})
endif()
endfunction(nv_test)
......@@ -387,8 +376,8 @@ function(hip_test TARGET_NAME)
endif()
add_executable(${TARGET_NAME} ${_cmake_options} ${_generated_files} ${_sources})
set_target_properties(${TARGET_NAME} PROPERTIES LINKER_LANGUAGE HIP)
target_link_libraries(${TARGET_NAME} ${hip_test_DEPS} paddle_gtest_main paddle_memory gtest gflags)
add_dependencies(${TARGET_NAME} ${hip_test_DEPS} paddle_gtest_main paddle_memory gtest gflags)
target_link_libraries(${TARGET_NAME} ${hip_test_DEPS} paddle_gtest_main memory gtest gflags)
add_dependencies(${TARGET_NAME} ${hip_test_DEPS} paddle_gtest_main memory gtest gflags)
add_test(${TARGET_NAME} ${TARGET_NAME})
endif()
endfunction(hip_test)
......@@ -561,9 +550,9 @@ function(py_test TARGET_NAME)
set(multiValueArgs SRCS DEPS ARGS ENVS)
cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
add_test(NAME ${TARGET_NAME}
COMMAND env PYTHONPATH=${PADDLE_PYTHON_BUILD_DIR}/lib-python ${py_test_ENVS}
COMMAND env PYTHONPATH=${PADDLE_BINARY_DIR}/python ${py_test_ENVS}
${PYTHON_EXECUTABLE} -u ${py_test_SRCS} ${py_test_ARGS}
WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR})
endif()
endfunction()
......@@ -587,6 +576,9 @@ function(grpc_library TARGET_NAME)
get_filename_component(PROTO_WE ${grpc_library_PROTO} NAME_WE)
get_filename_component(PROTO_PATH ${ABS_PROTO} PATH)
#FIXME(putcn): the follwoing line is supposed to generate *.pb.h and cc, but
# somehow it didn't. line 602 to 604 is to patching this. Leaving this here
# for now to enable dist CI.
protobuf_generate_cpp(grpc_proto_srcs grpc_proto_hdrs "${ABS_PROTO}")
set(grpc_grpc_srcs "${CMAKE_CURRENT_BINARY_DIR}/${PROTO_WE}.grpc.pb.cc")
set(grpc_grpc_hdrs "${CMAKE_CURRENT_BINARY_DIR}/${PROTO_WE}.grpc.pb.h")
......@@ -597,6 +589,9 @@ function(grpc_library TARGET_NAME)
COMMAND ${PROTOBUF_PROTOC_EXECUTABLE}
ARGS --grpc_out "${CMAKE_CURRENT_BINARY_DIR}" -I "${PROTO_PATH}"
--plugin=protoc-gen-grpc="${GRPC_CPP_PLUGIN}" "${ABS_PROTO}"
COMMAND ${PROTOBUF_PROTOC_EXECUTABLE}
ARGS --cpp_out "${CMAKE_CURRENT_BINARY_DIR}" -I "${PROTO_PATH}"
"${ABS_PROTO}"
DEPENDS "${ABS_PROTO}" ${PROTOBUF_PROTOC_EXECUTABLE} extern_grpc)
# FIXME(typhoonzero): grpc generated code do not generate virtual-dtor, mark it
......
cmake/inference_lib.cmake
浏览文件 @ d2ba05a6
# 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.
set_property(GLOBAL PROPERTY FLUID_MODULES "")
# find all fluid modules is used for paddle fluid static library
function(find_fluid_modules TARGET_NAME)
get_filename_component(__target_path ${TARGET_NAME} ABSOLUTE)
string(REGEX REPLACE "^${PADDLE_SOURCE_DIR}/" "" __target_path ${__target_path})
string(FIND "${__target_path}" "fluid" pos)
if(pos GREATER 1)
get_property(fluid_modules GLOBAL PROPERTY FLUID_MODULES)
......@@ -77,6 +92,23 @@ elseif (WITH_MKLML)
)
endif()
if(NOT MOBILE_INFERENCE AND NOT RPI)
set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/snappy")
copy(snappy_lib
SRCS ${SNAPPY_INCLUDE_DIR} ${SNAPPY_LIBRARIES}
DSTS ${dst_dir} ${dst_dir}/lib)
set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/snappystream")
copy(snappystream_lib
SRCS ${SNAPPYSTREAM_INCLUDE_DIR} ${SNAPPYSTREAM_LIBRARIES}
DSTS ${dst_dir} ${dst_dir}/lib)
set(dst_dir "${CMAKE_INSTALL_PREFIX}/third_party/install/zlib")
copy(zlib_lib
SRCS ${ZLIB_INCLUDE_DIR} ${ZLIB_LIBRARIES}
DSTS ${dst_dir} ${dst_dir}/lib)
endif()
# paddle fluid module
set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
set(dst_dir "${CMAKE_INSTALL_PREFIX}/paddle/fluid")
......
doc/CMakeLists.txt
浏览文件 @ d2ba05a6
add_custom_target(paddle_apis ALL
DEPENDS paddle_v2_apis paddle_fluid_apis)
add_custom_target(paddle_docs ALL
DEPENDS paddle_v2_docs paddle_v2_docs_cn
paddle_fluid_docs paddle_fluid_docs_cn
paddle_mobile_docs paddle_mobile_docs_cn)
add_subdirectory(v2)
add_subdirectory(fluid)
add_subdirectory(mobile)
doc/design/file_manager/README.md 已删除 100644 → 0
浏览文件 @ 10669f1f
# FileManager设计文档
## 目标
在本文档中，我们设计说明了名为FileManager系统，方便用户上传自己的训练数据以进行分布式训练
主要功能包括：
- 提供常用的命令行管理命令管理文件和目录
- 支持大文件的断点上传、下载
## 名词解释
- PFS：是`Paddlepaddle cloud File System`的缩写，是对用户文件存储空间的抽象，与之相对的是local filesystem。目前我们用CephFS来搭建。
- [CephFS](http://docs.ceph.com/docs/master/cephfs/)：一个POSIX兼容的文件系统。
- Chunk：逻辑划上文件分块的单位。
## 模块
### 架构图
<image src=./src/filemanager.png width=900>
### PFSClient
- 功能： 详细设计[link](./pfs/pfsclient.md)
- 提供用户管理文件的命令
- 需要可以跨平台执行
- 双向验证
PFSClient需要和Ingress之间做双向验证<sup>[tls](#tls)</sup>，所以用户需要首先在`cloud.paddlepaddle.org`上注册一下，申请用户空间，并且把系统生成的CA(certificate authority)、Key、CRT(CA signed certificate)下载到本地，然后才能使用PFSClient。
### [Ingress](https://kubernetes.io/docs/concepts/services-networking/ingress/)
- 功能：
提供七层协议的反向代理、基于粘性会话的负载均衡功能。
- 透传用户身份的办法
Ingress需要把PFSClient的身份信息传给PFSServer，配置的方法参考[link](http://www.integralist.co.uk/posts/clientcertauth.html#3)
### PFSServer
PFSServer提供RESTful API接口，接收处理PFSClient端的文件管理请求，并且把结果返回PFSClient端。
RESTful API
- /api/v1/files
- `GET /api/v1/files`: Get metadata of files or directories.
- `POST /api/v1/files`: Create files or directories.
- `PATCH /api/v1/files`: Update files or directories.
- `DELETE /api/v1/files`: Delete files or directories.
- /api/v1/file/chunks
- `GET /api/v1/storage/file/chunks`: Get chunks's metadata of a file.
- /api/v1/storage/files
- `GET /api/v1/storage/files`: Download files or directories.
- `POST /api/v1/storage/files`: Upload files or directories.
- /api/v1/storage/file/chunks
- `GET /api/v1/storage/file/chunks`: Download chunks's data.
- `POST /api/v1/storage/file/chunks`: Upload chunks's data.
## 文件传输优化
### 分块文件传输
用户文件可能是比较大的，上传到Cloud或者下载到本地的时间可能比较长，而且在传输的过程中也可能出现网络不稳定的情况。为了应对以上的问题，我们提出了Chunk的概念，一个Chunk由所在的文件偏移、数据、数据长度及校验值组成。文件的上传和下载都是通过对Chunk的操作来实现的。由于Chunk比较小（默认256K），完成一个传输动作完成的时间也比较短，不容易出错。PFSClient需要在传输完毕最后一个Chunk的时候检查destination文件的MD5值是否和source文件一致。
一个典型的Chunk如下所示：
```
type Chunk struct {
fileOffset int64
checksum uint32
len uint32
data []byte
}
```
### 生成sparse文件
当destination文件不存在或者大小和source文件不一致时，可以用[Fallocate](https://Go.org/pkg/syscall/#Fallocate)生成sparse文件，然后就可以并发写入多个Chunk。
### 覆盖不一致的部分
文件传输的的关键在于需要PFSClient端对比source和destination的文件Chunks的checksum是否保持一致，不一致的由PFSClient下载或者传输Chunk完成。这样已经传输成功的部分就不用重新传输了。
## 用户使用流程
参考[link](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/cluster_train/data_dispatch.md)
## 框架生成
用[swagger](https://github.com/swagger-api/swagger-codegen)生成PFSClient和PFSServer的框架部分，以便我们可以把更多的精力放到逻辑本身上。
## 参考文档
- <a name=tls></a>[TLS complete guide](https://github.com/k8sp/tls/blob/master/tls.md)
- [aws.s3](http://docs.aws.amazon.com/cli/latest/reference/s3/)
- [linux man document](https://linux.die.net/man/)
doc/design/file_manager/pfs/pfsclient.md 已删除 100644 → 0
浏览文件 @ 10669f1f
# PFSClient
## Description
The `pfs` command is a Command Line Interface to manage your files on PaddlePaddle Cloud
## Synopsis
```
paddle [options] pfs <subcommand> [parameters]
```
## Options
```
--profile (string)
Use a specific profile from your credential file.
--help (string)
Display more information about command
--version
Output version information and exit
--debug
Show detailed debugging log
--only-show-errors (boolean)
Only errors and warnings are displayed. All other output is suppressed.
```
## Path Arguments
When using a command, we need to specify path arguments. There are two path argument type: `localpath` and `pfspath`.
A `pfspath` begin with `/pfs`, eg: `/pfs/$DATACENTER/home/$USER/folder`.
[Here](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/cluster_train/data_dispatch.md#上传训练文件) is how to config datacenters.
## order of Path Arguments
Commonly, if there are two path arguments, the first is the source, and the second is the destination.
## Subcommonds
- rm - remove files or directories
```
Synopsis:
rm [-r] [-v] <PFSPath> ...
Options:
-r
Remove directories and their contents recursively
-v
Cause rm to be verbose, showing files after they are removed.
Examples:
paddle pfs rm /pfs/$DATACENTER/home/$USER/file
paddle pfs rm -r /pfs/$DATACENTER/home/$USER/folder
```
- mv - move (rename) files
```
Synopsis:
mv [-f | -n] [-v] <LocalPath> <PFSPath>
mv [-f | -n] [-v] <LocalPath> ... <PFSPath>
mv [-f | -n] [-v] <PFSPath> <LocalPath>
mv [-f | -n] [-v] <PFSPath> ... <LocalPath>
mv [-f | -n] [-v] <PFSPath> <PFSPath>
mv [-f | -n] [-v] <PFSPath> ... <PFSPath>
Options:
-f
Do not prompt for confirmation before overwriting the destination path. (The -f option overrides previous -n options.)
-n
Do not overwrite an existing file. (The -n option overrides previous -f options.)
-v
Cause mv to be verbose, showing files after they are moved.
Examples:
paddle pfs mv ./text1.txt /pfs/$DATACENTER/home/$USER/text1.txt
```
- cp - copy files or directories
```
Synopsis:
cp [-r] [-f | -n] [-v] [--preserve--links] <LocalPath> <PFSPath>
cp [-r] [-f | -n] [-v] [--preserve--links] <LocalPath> ... <PFSPath>
cp [-r] [-f | -n] [-v] [--preserve--links] <PFSPath> <LocalPath>
cp [-r] [-f | -n] [-v] [--preserve--links] <PFSPath> ... <LocalPath>
cp [-r] [-f | -n] [-v] [--preserve--links] <PFSPath> <PFSPath>
cp [-r] [-f | -n] [-v] [--preserve--links] <PFSPath> ... <PFSPath>
Options:
-r
Copy directories recursively
-f
Do not prompt for confirmation before overwriting the destination path. (The -f option overrides previous -n options.)
-n
Do not overwrite an existing file. (The -n option overrides previous -f options.)
-v
Cause cp to be verbose, showing files after they are copied.
--preserve--links
Reserve links when copy links
Examples:
paddle pfs cp ./file /pfs/$DATACENTER/home/$USER/file
paddle pfs cp /pfs/$DATACENTER/home/$USER/file ./file
```
- ls- list files
```
Synopsis:
ls [-r] <PFSPath> ...
Options:
-R
List directory(ies) recursively
Examples:
paddle pfs ls /pfs/$DATACENTER/home/$USER/file
paddle pfs ls /pfs/$DATACENTER/home/$USER/folder
```
- mkdir - mkdir directory(ies)
Create intermediate directory(ies) as required.
```
Synopsis:
mkdir <PFSPath> ...
Examples:
paddle pfs mkdir /pfs/$DATACENTER/home/$USER/folder
```
doc/fluid/CMakeLists.txt
浏览文件 @ d2ba05a6
......@@ -27,6 +27,8 @@ sphinx_add_target(paddle_fluid_docs
${CMAKE_CURRENT_SOURCE_DIR}
${SPHINX_HTML_DIR_EN})
add_dependencies(paddle_fluid_docs gen_proto_py paddle_python)
# configured documentation tools and intermediate build results
set(BINARY_BUILD_DIR_CN "${CMAKE_CURRENT_BINARY_DIR}/cn/_build")
......@@ -47,3 +49,7 @@ sphinx_add_target(paddle_fluid_docs_cn
${SPHINX_CACHE_DIR_CN}
${CMAKE_CURRENT_SOURCE_DIR}
${SPHINX_HTML_DIR_CN})
add_dependencies(paddle_fluid_docs_cn gen_proto_py paddle_python)
add_subdirectory(api)
doc/fluid/api/CMakeLists.txt 0 → 100644
浏览文件 @ d2ba05a6
# configured documentation tools and intermediate build results
set(BINARY_BUILD_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/_build")
# Sphinx cache with pickled ReST documents
set(SPHINX_CACHE_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/_doctrees")
# HTML output director
set(SPHINX_HTML_DIR_EN "${CMAKE_CURRENT_BINARY_DIR}/en/html")
configure_file(
"${CMAKE_CURRENT_SOURCE_DIR}/../../templates/conf.py.en.in"
"${BINARY_BUILD_DIR_EN}/conf.py"
@ONLY)
sphinx_add_target(paddle_fluid_apis
html
${BINARY_BUILD_DIR_EN}
${SPHINX_CACHE_DIR_EN}
${CMAKE_CURRENT_SOURCE_DIR}
${SPHINX_HTML_DIR_EN})
add_dependencies(paddle_fluid_apis gen_proto_py framework_py_proto copy_paddle_pybind paddle_python)
doc/v2/api/fluid/layers.rst → doc/fluid/api/layers.rst
浏览文件 @ d2ba05a6
......@@ -473,6 +473,12 @@ multiplex
.. autofunction:: paddle.fluid.layers.multiplex
:noindex:
label_smooth
------------
.. autofunction:: paddle.fluid.layers.label_smooth
:noindex:
ops
===
......@@ -494,6 +500,12 @@ reshape
.. autofunction:: paddle.fluid.layers.reshape
:noindex:
pad
---
.. autofunction:: paddle.fluid.layers.pad
:noindex:
scale
-----
......
doc/fluid/build_and_install/build_from_source_cn.rst 0 → 120000
浏览文件 @ d2ba05a6
../../v2/build_and_install/build_from_source_cn.rst
\ No newline at end of file
doc/fluid/build_and_install/build_from_source_en.rst 0 → 120000
浏览文件 @ d2ba05a6
../../v2/build_and_install/build_from_source_en.rst
\ No newline at end of file
doc/fluid/build_and_install/docker_install_cn.rst 0 → 120000
浏览文件 @ d2ba05a6
../../v2/build_and_install/docker_install_cn.rst
\ No newline at end of file
doc/fluid/build_and_install/docker_install_en.rst 0 → 120000
浏览文件 @ d2ba05a6
../../v2/build_and_install/docker_install_en.rst
\ No newline at end of file
doc/fluid/build_and_install/index_cn.rst 0 → 120000
浏览文件 @ d2ba05a6
../../v2/build_and_install/index_cn.rst
\ No newline at end of file
doc/fluid/build_and_install/index_en.rst 0 → 120000
浏览文件 @ d2ba05a6
../../v2/build_and_install/index_en.rst
\ No newline at end of file
doc/fluid/build_and_install/pip_install_cn.rst 0 → 120000
浏览文件 @ d2ba05a6
../../v2/build_and_install/pip_install_cn.rst
\ No newline at end of file
doc/fluid/build_and_install/pip_install_en.rst 0 → 120000
浏览文件 @ d2ba05a6
../../v2/build_and_install/pip_install_en.rst
\ No newline at end of file
doc/fluid/design/algorithm/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
梯度更新算法
------------
.. toctree::
:maxdepth: 1
parameter_average.md
doc/fluid/design/algorithm/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
Gradient Update Algorithm
--------------------------------------
.. toctree::
:maxdepth: 1
parameter_average.md
doc/fluid/design/algorithm/parameter_average.md
浏览文件 @ d2ba05a6
......@@ -5,9 +5,11 @@ In a large scale machine learning setup where the size of the training data is h
Polyak and Juditsky (1992) showed that the test performance of simple average of parameters obtained by Stochastic Gradient Descent (SGD) is as good as that of parameter values that are obtained by training the model over and over again, over the training dataset.
Hence, to accelerate the speed of Stochastic Gradient Descent, Averaged Stochastic Gradient Descent (ASGD) was proposed in Polyak and Juditsky (1992). For ASGD, the running average of parameters obtained by SGD, is used as the estimator for <img src="./images/theta_star.gif"/><br/> . The averaging is done as follows:
Hence, to accelerate the speed of Stochastic Gradient Descent, Averaged Stochastic Gradient Descent (ASGD) was proposed in Polyak and Juditsky (1992). For ASGD, the running average of parameters obtained by SGD, is used as the estimator for <img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/fluid/images/theta_star.gif"/><br/> . The averaging is done as follows:
<img src="./images/asgd.gif" align="center"/><br/>
<p align="center">
<img src="https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/doc/fluid/images/asgd.gif"><br />
</p>
We propose averaging for any optimizer similar to how ASGD performs it, as mentioned above.
......
doc/fluid/design/concepts/README.md
浏览文件 @ d2ba05a6
......@@ -2,15 +2,37 @@ A few months ago when we were trying to replace CMake with Bazel, @emailweixu su
Here are some initial thoughts. Your comments are welcome!
### Required CMake Function
# Required CMake Function
I think we need only the following few CMake functions to make a project description mean and clean:
| C++ | CUDA C++ | Go |
|---|---|---|
| cc_library | nv_library | go_library |
| cc_binary | nv_binary | go_binary |
| cc_test | nv_test | go_test |
<table>
<thead>
<tr>
<th>C++</th>
<th>CUDA C++</th>
<th>Go</th>
</tr>
</thead>
<tbody>
<tr>
<td>cc_library </td>
<td>nv_library </td>
<td>go_library </td>
</tr>
<tr>
<td>cc_binary </td>
<td>nv_binary </td>
<td>go_binary </td>
</tr>
<tr>
<td> cc_test </td>
<td> nv_test </td>
<td> go_test </td>
</tr>
</tbody>
</table>
- The `_library` functions generate .a files from source code.
- The `_binary` functions generate executable binary files.
......@@ -25,7 +47,7 @@ Also,
- to describe external dependencies, we need `external_library`.
- to build shared libraries, we need `shared_library`.
### An Example Project
## An Example Project
Suppose that we have aforementioned functions defined in our `/cmake` directory. The following example `CMakeLists.txt` describes a project including the following source files:
......@@ -102,11 +124,11 @@ shared_library(api
```
### Implementation
## Implementation
As above example CMakeLists.txt executes, each function invocation adds "nodes" to a dependency graph. It also use this graph to generate CMake commands including `add_executable`, `add_dependencies`, `target_link_libraries`, and `add_test`.
### Using Package Manager For Go
## Using Package Manager For Go
Building Go binaries and libraries need to satisfy their dependencies, generally
we can do `go get ./...` to download and compile all external dependencies. The
......@@ -122,7 +144,7 @@ problems are:
at many cloud file hosting, so users what to compile paddle by themselves can
download this "vendor" package from a mirror site.
#### Choose A Suitable Tool
### Choose A Suitable Tool
As mentioned by @wangkuiyi, [Here](https://github.com/golang/go/wiki/PackageManagementTools)
list dozens of Go package managers. We choose the tool using following principles:
......@@ -140,7 +162,7 @@ management tool has been started at: https://github.com/golang/dep to resolve
such problems, but it's currently at Alpha stage. So the best choice now is
glide obviously.
#### Manage Go Packages
### Manage Go Packages
- Dependencies: `go/glide.yaml` will store the dependencies and their versions which
is directly imported by paddle. `go/glide.lock` will store all dependencies recursively
......
doc/fluid/design/concepts/block.md
浏览文件 @ d2ba05a6
......@@ -14,11 +14,29 @@ In programming languages, a block is a pair of curly braces that includes local
Blocks work with control flow structures like `if`, `else`, and `for`, which have equivalents in deep learning:
| programming languages | PaddlePaddle |
|-----------------------|-----------------------|
| for, while loop | RNN, WhileOp |
| if, if-else, switch | IfElseOp, SwitchOp |
| sequential execution | a sequence of layers |
<table>
<thead>
<tr>
<th>programming languages</th>
<th>PaddlePaddle</th>
</tr>
</thead>
<tbody>
<tr>
<td>for, while loop </td>
<td>RNN, WhileOp </td>
</tr>
<tr>
<td>if, if-else, switch </td>
<td>IfElseOp, SwitchOp </td>
</tr>
<tr>
<td>sequential execution </td>
<td>a sequence of layers </td>
</tr>
</tbody>
</table>
A key difference is that a C++ program describes a one pass computation, whereas a deep learning program describes both the forward and backward passes.
......@@ -26,12 +44,33 @@ A key difference is that a C++ program describes a one pass computation, whereas
The existence of the backward pass makes the execution of a block of PaddlePaddle different from traditional programs:
| programming languages | PaddlePaddle |
|-----------------------|---------------------------------|
| stack | scope hierarchy |
| stack frame | scope |
| push at entering block| push at entering block |
| pop at leaving block | destroy when minibatch completes|
<table>
<thead>
<tr>
<th>programming languages</th>
<th>PaddlePaddle</th>
</tr>
</thead>
<tbody>
<tr>
<td>stack </td>
<td>scope hierarchy </td>
</tr>
<tr>
<td>stack frame </td>
<td>scope </td>
</tr>
<tr>
<td>push at entering block </td>
<td>push at entering block </td>
</tr>
<tr>
<td>pop at leaving block </td>
<td>destroy when minibatch completes </td>
</tr>
</tbody>
</table>
1. In traditional programs:
......
doc/fluid/design/concepts/cpp_data_feeding.md
浏览文件 @ d2ba05a6
......@@ -113,7 +113,7 @@ To solve this problem, we introduce `ReaderHolder` as a wrapper. It acts as an e
To create and invoke readers, some new ops are introduced:
### CreateReaderOp
### Operators That Create Readers
Each reader has its creation op. File readers' creation ops have no input and yield the created file reader as its output. Decorated readers' creation ops take the underlying readers as inputs and then yield new decorated readers.
......@@ -153,19 +153,52 @@ double_buffer_reader = create_double_buffer_op(batch_reader)
The forwarding ops of the corresponding `main_program` would be like this:
```
while_op {
not_completed = true
pass_count = 0
while_op(not_completed) {
has_next = has_next_op(double_buffer_reader)
if_else_op(has_next) {
batch_data = read_op(double_buffer_reader)
... (subsequent training ops)
} else {
reset_op(double_buffer_reader)
increase_op(pass_count)
not_completed = less_than_op(pass_count, reqiured_pass_num)
}
}
```
Two important considerations for these programs are as follows:
A few important considerations for these programs are as follows:
1. The multiple\_reader is the batch\_reader's underlying reader, and the batch\_reader is the double\_buffer\_reader's underlying reader. `read_op`, `has_next_op` and other reader related ops will only invoke the top-most reader. In this case, it's the double\_buffer\_reader.
1. `not_completed`, `pass_count` and other variables shown above are all Fluid Variables.
2. All readers exist in both `startup_program` and `main_program`. And they are persistable.
2. The multiple\_reader is the batch\_reader's underlying reader, and the batch\_reader is the double\_buffer\_reader's underlying reader. `read_op`, `has_next_op` and other reader related ops will only invoke the top-most reader. In this case, it's the double\_buffer\_reader.
3. All readers exist in both `startup_program` and `main_program`. And they are persistable.
### Simplify Configuration by MultiPassReader
The Program configuration mentioned above is complicated. Users need to be very familiar to concepts of Program and Block to prevent making mistakes in their code. To make the usage of C++ readers more friendly to new users, we introduce `MultiPassReader`.
`MultiPassReader` is a decorated reader. A multi-pass reader is used to continuously yield data for several training passes. It takes the number of passes to run as one of its attributes('pass_num') and maintains a counter to record how many passes it has completed. Each time its underlying reader reaches the EOF, the multi-pass reader checks whether it has completed the training of given number of pass. If not, the underlying reader will be re-initialized and starts a new pass automatically. Before completing the whole training, the return of MultiPassReader's `HasNext()` will always be `true`.
With `MultiPassReader`, the startup program would be like this:
```
multiple_reader = open_files_op(...)
batch_reader = create_batch_reader_op(multiple_reader)
multi_pass_reader = create_multi_pass_reader_op(batch_reader)
double_buffer_reader = create_double_buffer_op(multi_pass_reader)
... (other initializers)
```
The forwarding part of the corresponding `main_program` would be like this:
```
not_completed = true
while_op(not_completed) {
batch_data = read_op(double_buffer_reader)
... (subsequent training ops)
not_completed = has_next_op(double_buffer_reader)
}
```
doc/fluid/design/concepts/functions_operators_layers.md
浏览文件 @ d2ba05a6
......@@ -86,12 +86,40 @@ def layer.fc(X):
We'd like to have Python bindings to operators in package `paddle.operator`, and Python compositions of operators in package `paddle.layer`. So we have the following concepts in above illustrative example:
| C++ functions/functors | mul | add | | |
|------------------------|--------------|--------------|-------------|----------|
| C++ operator class | mulOp | addOp | FCOp | |
| Python binding | operator.mul | operator.add | operator.fc | |
| Python function | | | | layer.fc |
<table>
<thead>
<tr>
<th>C++ functions/functors</th>
<th>mul</th>
<th>add</th>
<th></th>
<th></th>
</tr>
</thead>
<tbody>
<tr>
<td>C++ operator class </td>
<td>mulOp</td>
<td>addOp </td>
<td>FCOp </td>
<td></td>
</tr>
<tr>
<td>Python binding </td>
<td>operator.mul</td>
<td> operator.add </td>
<td>operator.fc </td>
<td></td>
</tr>
<tr>
<td>Python function </td>
<td></td>
<td></td>
<td> </td>
<td>layer.fc</td>
</tr>
</tbody>
</table>
This is how we differentiate layer and operators in PaddlePaddle:
......
doc/fluid/design/concepts/images/parallel_executor_overview.dot 0 → 100644
浏览文件 @ d2ba05a6
digraph G {
subgraph cluster_init {
label="Initialization"
startup_program [label="startup", shape=box]
node_w_g0 [label="W\nGPU0"]
startup_program -> node_w_g0 [label="Initialize"]
node_w_g1 [label="W\nGPU1"]
node_w_g0 -> node_w_g1 [label="broadcast"]
}
subgraph cluster_train {
label="forward_backward"
subgraph cluster_gpu0 {
label="GPU0"
fc_0 [label="fc\nGPU0", shape=box]
hidden_0 [label="hidden\nGPU0"]
node_w_g0 -> fc_0
fc_0 -> hidden_0
loss0 [label="loss\nGPU0"]
hidden_0 -> loss0 [label="many ops omitted"]
scale_loss_0 [label="scale_loss_gradient\nGPU0", shape=box]
loss_g0 [label="loss_grad\nGPU0"]
scale_loss_0->loss_g0
fc_g_0 [label="w_grad\nGPU0", shape=box]
loss0 -> fc_g_0
loss_g0 -> fc_g_0
hidden_0 -> fc_g_0
}
subgraph cluster_gpu1 {
label="GPU1"
fc_1 [label="fc\nGPU1", shape=box]
hidden_1 [label="hidden\nGPU1"]
node_w_g1 -> fc_1
fc_1 -> hidden_1
loss1 [label="loss\nGPU1"]
hidden_1 -> loss1 [label="many ops omitted"]
scale_loss_1 [label="scale_loss_gradient\nGPU1", shape=box]
loss_g1 [label="loss_grad\nGPU1"]
scale_loss_1->loss_g1
fc_g_1 [label="w_grad\nGPU1", shape=box]
loss1 -> fc_g_1
loss_g1 -> fc_g_1
hidden_1 -> fc_g_1
}
}
all_reduce_w [label="Merge Gradients(AllReduce)", shape=box]
fc_g_0 -> all_reduce_w
fc_g_1 -> all_reduce_w
fc_g_0_merged [label="w_grad\nMerged\nGPU0"]
fc_g_1_merged [label="w_grad\nMerged\nGPU1"]
all_reduce_w -> fc_g_0_merged
all_reduce_w -> fc_g_1_merged
subgraph cluster_optimization {
label="Optimization"
subgraph cluster_opt_gpu0 {
label="GPU0"
sgd_0 [label="SGD Op\nGPU0", shape=box]
fc_g_0_merged -> sgd_0
node_w_g0 -> sgd_0
optimized_w_0 [label="Optimized W\nGPU0"]
sgd_0 -> optimized_w_0
}
subgraph cluster_opt_gpu1 {
label="GPU1"
sgd_1 [label="SGD Op\nGPU1", shape=box]
fc_g_1_merged -> sgd_1
node_w_g1 -> sgd_1
optimized_w_1 [label="Optimized W\nGPU0"]
sgd_1 -> optimized_w_1
}
}
}
doc/fluid/design/concepts/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
核心概念
-------------
.. toctree::
:maxdepth: 1
README.md
cpp_data_feeding.md
functions_operators_layers.md
program.md
variable.md
var_desc.md
tensor.md
tensor_array.md
lod_tensor.md
block.md
scope.md
executor.md
parallel_executor.md
doc/fluid/design/concepts/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
Core Concepts
--------------------------------------
.. toctree::
:maxdepth: 1
README.md
cpp_data_feeding.md
functions_operators_layers.md
program.md
variable.md
var_desc.md
tensor.md
tensor_array.md
lod_tensor.md
block.md
scope.md
executor.md
parallel_executor.md
doc/fluid/design/concepts/lod_tensor.md
浏览文件 @ d2ba05a6
......@@ -2,12 +2,38 @@
Like other deep learning systems, PaddlePaddle supports training models from sequence data. Also, like other systems, PaddlePaddle represent a mini-batch of sequences as a Tensor. What is different is that PaddlePaddle doesn't require all sequences in a mini-batch to be of the same length. Thus no need for padding zeros.
| | TensorFlow | PaddlePaddle |
|-----------------------|------------|--------------|
| RNN | Support | Support |
| recursive RNN | Support | Support |
| padding zeros | Must | No need |
| blob data type | Tensor | LoDTensor |
<table>
<thead>
<tr>
<th></th>
<th>TensorFlow</th>
<th>PaddlePaddle</th>
</tr>
</thead>
<tbody>
<tr>
<td>RNN </td>
<td>Support </td>
<td>Support </td>
</tr>
<tr>
<td>recursive RNN </td>
<td>Support </td>
<td>Support </td>
</tr>
<tr>
<td>padding zeros </td>
<td> Must </td>
<td>No need </td>
</tr>
<tr>
<td> blob data type </td>
<td> Tensor</td>
<td> LoDTensor </td>
</tr>
</tbody>
</table>
PaddlePaddle achieves this flexibility by passing through a new data type, *LoD Tensor*, which is a Tensor attached with segmentation index known as *LoD*, between operators. The LoD index doesn't only segment a tensor, but also recursively segments sub-sequences. This document presents the design of LoD and LoDTensor.
......
doc/fluid/design/concepts/parallel_executor.md 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/concepts/scope.md
浏览文件 @ d2ba05a6
......@@ -30,7 +30,7 @@ Scope is an association of a name to variable. All variables belong to `Scope`.
Variable can not belong to many scopes. If you want to use variables from parent scope, you can use `parent scope`.
1. Scope should destruct all Variables inside it when itself is destructed. User can never store `Variable` pointer somewhere else.
1. Scope should destruct all Variables inside it when itself is destructed. User can never store `Variable` pointer somewhere else.
Because Variable can only be got from Scope. When destroying Scope, we also need to destroy all the Variables in it. If user store `Variable` pointer to private data member or some global variable, the pointer will be an invalid pointer when associated `Scope` is destroyed.
......@@ -78,7 +78,7 @@ In `Scope` class, there is a private data member called `parent_`. `parent_` is
A local scope is very useful when we implement Recurrent Neural Network. Each timestep of an RNN should be a `Net`. Each `Net` of timestep (`StepNet` for short) should use an independent local scope. Just like variables in a while loop is inside a local scope in programming languages. By using a single `StepNet` and changing local scope, we can implement an RNN easily.
# Interface Design
## Interface Design
```cpp
class Variable {
......
doc/fluid/design/concurrent/channel.md 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/concurrent/csp.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/concurrent/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
并发编程
------------
.. toctree::
:maxdepth: 1
concurrent_programming.md
parallel_do.md
doc/fluid/design/concurrent/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
Concurrent Programming
-------------------------
.. toctree::
:maxdepth: 1
concurrent_programming.md
parallel_do.md
doc/fluid/design/data_type/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
数据类型
------------
.. toctree::
:maxdepth: 1
float16.md
doc/fluid/design/data_type/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dist_train/README.md 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dist_train/async_update.md 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dist_train/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dist_train/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dist_train/src/async_update.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dist_train/src/async_update.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dynamic_rnn/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dynamic_rnn/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/dynamic_rnn/rnn.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/execution/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/execution/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/index_cn.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/index_en.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/interface/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/interface/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/memory/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/memory/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/modules/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/modules/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/motivation/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/motivation/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/muti_devices/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/muti_devices/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/network/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/network/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/design/others/gan_api.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/api_doc_std_cn.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/api_doc_std_en.md 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/index_cn.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/index_en.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/name_convention.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/new_op_cn.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/new_op_en.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/op_markdown_format.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/releasing_process_en.md 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/use_eigen_cn.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/dev/use_eigen_en.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/getstarted/concepts/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/getstarted/concepts/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/getstarted/index_cn.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/getstarted/index_en.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/getstarted/quickstart_cn.rst 0 → 120000
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/getstarted/quickstart_en.rst 0 → 120000
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/howto/index_cn.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/howto/index_en.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/howto/optimization/benchmark/README.md 0 → 120000
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/howto/optimization/benchmark/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/howto/optimization/benchmark/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/howto/optimization/benchmark/vgg16/README.md 0 → 120000
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/howto/optimization/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/howto/optimization/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/2_level_rnn.dot 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/2_level_rnn.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/asgd.gif 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/batch_norm_fork.dot 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/batch_norm_fork.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/batch_norm_op_kernel.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/beam_search.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/ci_build_whl.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/compiler.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/compiler.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/control_flow_graph.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/dataflow_equations.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/dcgan.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/deep_learning.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/dist-graph.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/dist-graph.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/distributed_architecture.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/distributed_architecture.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/ds2_network.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/feed_forward.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/feed_forward_regularized.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/fluid-compiler.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/fluid-compiler.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/graph_construction_example.bash 0 → 100755
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/graph_construction_example.dot 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/graph_construction_example_all.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/l1_regularization.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/l2_regularization.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/local-graph.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/local-graph.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/local_architecture.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/local_architecture.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/lookup_table.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/lookup_table_training.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/loss_equation.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/multi-threads.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/multi-threads@3x.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/multigpu_allreduce.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/multigpu_allreduce.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/multigpu_before_convert.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/multigpu_before_convert.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/multiple_reader.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/paddle-compile.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/paddle-compile.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/pprof_1.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/pprof_2.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/profiler.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/readers.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/remote_executor.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/remote_executor.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/rnn.dot 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/rnn.jpg 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/rnn.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/rnn_2level_data.dot 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/rnn_2level_data.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/single-thread@3x.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/sparse_update.graffle 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/sparse_update.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/test.dot 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/test.dot.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/theta_star.gif 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/timeline.jpeg 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/images/tracing.jpeg 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/index_cn.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/fluid/index_en.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/mobile/CMakeLists.txt 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/mobile/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/mobile/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/templates/conf.py.cn.in
浏览文件 @ d2ba05a6
此差异已折叠。
doc/templates/conf.py.en.in
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/api/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/design/interface/index_cn.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/design/interface/index_en.rst 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/design/mkl/mkldnn.md
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/dev/write_docs_cn.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/dev/write_docs_en.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/faq/cluster/index_en.rst
浏览文件 @ d2ba05a6
此差异已折叠。
doc/v2/faq/model/index_en.rst
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/.gitignore
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/api/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/api/test/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/cuda/include/hl_cnn.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/cuda/src/hl_cuda_cnn.cu
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/backward.cc 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/framework/backward_test.cc 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/framework/block_desc.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/channel.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/CMakeLists.txt 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/broadcast_op_handle.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/broadcast_op_handle.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/computation_op_handle.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/computation_op_handle.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/fetch_op_handle.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/fetch_op_handle.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/gather_op_handle.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/gather_op_handle.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/gather_op_handle_test.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/op_handle_base.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/op_handle_base.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/send_op_handle.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/send_op_handle.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/ssa_graph.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/ssa_graph.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/ssa_graph_builder.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/ssa_graph_builder.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/ssa_graph_executor.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/ssa_graph_executor.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/var_handle.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/details/var_handle.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/executor.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/executor.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/init.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/init.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/init_test.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/lod_tensor.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/operator.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/operator.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/parallel_executor.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/reader.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/reader.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/scope.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/scope.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/tensor.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/threadpool.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/framework/tuple.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/inference/io.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/inference/io.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/inference/tensorrt/CMakeLists.txt 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/inference/tensorrt/test_tensorrt.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/inference/tests/test_multi_thread_helper.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/memory/.clang-format 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/memory/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/memory/malloc.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/memory/memcpy.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/memory/memory.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/memory/pinned_memory_test.cu 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/.clang-format 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/operators/auc_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/auc_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/concat_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/cond_op.cc 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/operators/cond_op.h 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/operators/conv_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/crop_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/detail/grpc_server_test.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/dropout_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/fc_mkldnn_op.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/fc_op.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/go_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/gru_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/gru_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/increment_op.cu 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/increment_op.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/lrn_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/lrn_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/lstm_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/lstm_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/lstmp_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/lstmp_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/matmul_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/matmul_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/maxout_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/minus_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/mul_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/net_op.cc 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/operators/net_op.h 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/operators/net_op_test.cc 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/operators/pad_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/pool_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/prefetch_op.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/prelu_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/prior_box_op.cu 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/read_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/recv_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/reshape_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/scale_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/scatter_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/send_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/sgd_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/sgd_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/split_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/split_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/spp_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/spp_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/sum_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/sum_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/top_k_op.cu
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/top_k_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/unpool_op.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/unpool_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/operators/warpctc_op.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/.clang-format 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/platform/cpu_info.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/cpu_info.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/cuda_helper.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/dynload/tensorrt.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/dynload/tensorrt.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/enforce.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/float16.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/gpu_info.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/gpu_info.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/nccl_helper.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/place.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/place.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/profiler.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/profiler.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/transform.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/platform/variant.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/.clang-format 已删除 100644 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/pybind/.gitignore 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/const_value.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/const_value.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/exception.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/exception.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/protobuf.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/protobuf.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/pybind.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/recordio.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/recordio.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/tensor_py.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/pybind/tensor_py_test.cc 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/chunk.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/chunk.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/chunk_test.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/header.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/header.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/scanner.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/scanner.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/writer.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/recordio/writer.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/string/.clang-format 已删除 120000 → 0
浏览文件 @ 10669f1f
此差异已折叠。
paddle/fluid/string/piece.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/string/printf.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/fluid/string/printf_test.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/gserver/layers/UpsampleLayer.cpp 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/gserver/layers/UpsampleLayer.h 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/gserver/tests/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/gserver/tests/test_Upsample.cpp 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/math/Matrix.cpp
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/math/Matrix.h
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/scripts/docker/build.sh
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/scripts/submit_local.sh.in
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/scripts/travis/build_doc.sh
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/testing/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/testing/paddle_gtest_main.cc
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/trainer/tests/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/utils/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/utils/DynamicLoader.cpp
浏览文件 @ d2ba05a6
此差异已折叠。
paddle/utils/DynamicLoader.h
浏览文件 @ d2ba05a6
此差异已折叠。
proto/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
proto/ModelConfig.proto
浏览文件 @ d2ba05a6
此差异已折叠。
python/.gitignore
浏览文件 @ d2ba05a6
此差异已折叠。
python/CMakeLists.txt
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/.gitignore 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/__init__.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/batch.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/__init__.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/cifar.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/common.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/conll05.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/flowers.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/image.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/imdb.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/imikolov.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/mnist.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/movielens.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/mq2007.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/sentiment.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/CMakeLists.txt 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/cat.jpg 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/cifar_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/common_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/flowers_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/imdb_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/imikolov_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/mnist_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/mq2007_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/test_sentiment.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/voc2012_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/tests/wmt16_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/uci_housing.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/voc2012.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/wmt14.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/dataset/wmt16.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/__init__.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/average.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/debuger.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/evaluator.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/executor.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/framework.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/graphviz.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/initializer.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/layers/io.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/layers/nn.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/layers/ops.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/metrics.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/nets.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/optimizer.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/parallel_executor.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/param_attr.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/tests/unittests/test_fc_op.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/fluid/tests/unittests/test_target_assign_op.py 100755 → 100644
浏览文件 @ d2ba05a6
文件模式从 100755 更改为 100644
python/paddle/reader/__init__.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/creator.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/decorator.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/tests/CMakeLists.txt 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/tests/__init__.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/tests/creator_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/tests/decorator_test.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/tests/test_data_creator.txt 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/tests/test_reader_recordio.dat 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/reader/tests/test_recordio_creator.dat 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/v2/inference.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/v2/layer.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/paddle/v2/reader/__init__.py
浏览文件 @ d2ba05a6
此差异已折叠。
python/setup.py.in
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/README.md 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/client/Dockerfile 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/client/cluster_launcher.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/client/requirements.txt 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/diagram.png 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/server/Dockerfile 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/server/cluster_master.py 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/server/pserver.sh.template 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/server/requirements.txt 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/aws_benchmarking/server/trainer.sh.template 0 → 100644
浏览文件 @ d2ba05a6
此差异已折叠。
tools/codestyle/cpplint_pre_commit.hook 0 → 100755
浏览文件 @ d2ba05a6
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