Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
wzpei9
Paddle
提交
44cb70c0
P
Paddle
项目概览
wzpei9
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
1
Issue
1
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
44cb70c0
编写于
11月 07, 2018
作者:
T
tensor-tang
浏览文件
操作
浏览文件
下载
差异文件
Merge remote-tracking branch 'ups/develop' into fix/mac
上级
c9730d33
2466ca13
变更
16
隐藏空白更改
内联
并排
Showing
16 changed file
with
177 addition
and
126 deletion
+177
-126
paddle/fluid/framework/details/execution_strategy.h
paddle/fluid/framework/details/execution_strategy.h
+2
-0
paddle/fluid/framework/details/fast_threaded_ssa_graph_executor.cc
...uid/framework/details/fast_threaded_ssa_graph_executor.cc
+3
-1
paddle/fluid/framework/details/threaded_ssa_graph_executor.cc
...le/fluid/framework/details/threaded_ssa_graph_executor.cc
+3
-1
paddle/fluid/framework/details/threaded_ssa_graph_executor.h
paddle/fluid/framework/details/threaded_ssa_graph_executor.h
+1
-1
paddle/fluid/framework/parallel_executor.cc
paddle/fluid/framework/parallel_executor.cc
+12
-11
paddle/fluid/inference/CMakeLists.txt
paddle/fluid/inference/CMakeLists.txt
+1
-1
paddle/fluid/inference/tests/api/CMakeLists.txt
paddle/fluid/inference/tests/api/CMakeLists.txt
+16
-7
paddle/fluid/inference/tests/api/analyzer_resnet50_tester.cc
paddle/fluid/inference/tests/api/analyzer_resnet50_tester.cc
+2
-28
paddle/fluid/inference/tests/api/tester_helper.h
paddle/fluid/inference/tests/api/tester_helper.h
+29
-0
paddle/fluid/inference/tests/test.cmake
paddle/fluid/inference/tests/test.cmake
+0
-0
paddle/fluid/inference/tests/test_helper.h
paddle/fluid/inference/tests/test_helper.h
+8
-18
paddle/fluid/pybind/pybind.cc
paddle/fluid/pybind/pybind.cc
+6
-1
python/paddle/fluid/layers/io.py
python/paddle/fluid/layers/io.py
+1
-1
python/paddle/fluid/tests/unittests/test_parallel_executor_dry_run.py
...e/fluid/tests/unittests/test_parallel_executor_dry_run.py
+80
-0
python/paddle/fluid/tests/unittests/test_parallel_executor_mnist.py
...dle/fluid/tests/unittests/test_parallel_executor_mnist.py
+9
-52
python/setup.py.in
python/setup.py.in
+4
-4
未找到文件。
paddle/fluid/framework/details/execution_strategy.h
浏览文件 @
44cb70c0
...
...
@@ -13,6 +13,7 @@
// limitations under the License.
#pragma once
#include <cstddef> // for size_t
namespace
paddle
{
namespace
framework
{
...
...
@@ -26,6 +27,7 @@ struct ExecutionStrategy {
bool
allow_op_delay_
{
false
};
size_t
num_iteration_per_drop_scope_
{
100
};
ExecutorType
type_
{
kDefault
};
bool
dry_run_
{
false
};
};
}
// namespace details
...
...
paddle/fluid/framework/details/fast_threaded_ssa_graph_executor.cc
浏览文件 @
44cb70c0
...
...
@@ -128,7 +128,9 @@ void FastThreadedSSAGraphExecutor::RunOpAsync(
size_t
complete
=
0
;
while
(
op_to_run
!=
nullptr
)
{
try
{
op_to_run
->
Run
(
strategy_
.
use_cuda_
);
if
(
LIKELY
(
!
strategy_
.
dry_run_
))
{
op_to_run
->
Run
(
strategy_
.
use_cuda_
);
}
++
complete
;
}
catch
(...)
{
exception_
.
Catch
(
std
::
current_exception
());
...
...
paddle/fluid/framework/details/threaded_ssa_graph_executor.cc
浏览文件 @
44cb70c0
...
...
@@ -211,7 +211,9 @@ void ThreadedSSAGraphExecutor::RunOp(
if
(
VLOG_IS_ON
(
10
))
{
VLOG
(
10
)
<<
op
<<
" "
<<
op
->
Name
()
<<
" : "
<<
op
->
DebugString
();
}
op
->
Run
(
strategy_
.
use_cuda_
);
if
(
LIKELY
(
!
strategy_
.
dry_run_
))
{
op
->
Run
(
strategy_
.
use_cuda_
);
}
VLOG
(
10
)
<<
op
<<
" "
<<
op
->
Name
()
<<
" Done "
;
running_ops_
--
;
ready_var_q
->
Extend
(
op
->
Outputs
());
...
...
paddle/fluid/framework/details/threaded_ssa_graph_executor.h
浏览文件 @
44cb70c0
...
...
@@ -48,7 +48,7 @@ class ThreadedSSAGraphExecutor : public SSAGraphExecutor {
// Use topological sort algorithm
FeedFetchList
Run
(
const
std
::
vector
<
std
::
string
>
&
fetch_tensors
)
override
;
~
ThreadedSSAGraphExecutor
()
{}
~
ThreadedSSAGraphExecutor
()
final
=
default
;
private:
void
RunOp
(
const
std
::
shared_ptr
<
BlockingQueue
<
VarHandleBase
*>>
&
ready_var_q
,
...
...
paddle/fluid/framework/parallel_executor.cc
浏览文件 @
44cb70c0
...
...
@@ -38,9 +38,20 @@ class ParallelExecutorPrivate {
explicit
ParallelExecutorPrivate
(
const
std
::
vector
<
platform
::
Place
>
&
places
)
:
places_
(
places
)
{}
~
ParallelExecutorPrivate
()
{
if
(
own_local_scope_
)
{
for
(
size_t
i
=
1
;
i
<
local_scopes_
.
size
();
++
i
)
{
// Skip the first scope, since it is the global scope.
Scope
*
local_scope
=
local_scopes_
[
i
];
if
(
global_scope_
->
HasKid
(
local_scope
))
{
global_scope_
->
DeleteScope
(
local_scope
);
}
}
}
}
std
::
vector
<
platform
::
Place
>
places_
;
std
::
vector
<
Scope
*>
local_scopes_
;
Scope
*
global_scope_
;
Scope
*
global_scope_
;
// not owned
std
::
unique_ptr
<
details
::
SSAGraphExecutor
>
executor_
;
#ifdef PADDLE_WITH_CUDA
...
...
@@ -306,16 +317,6 @@ ParallelExecutor::~ParallelExecutor() {
for
(
auto
&
p
:
member_
->
places_
)
{
platform
::
DeviceContextPool
::
Instance
().
Get
(
p
)
->
Wait
();
}
if
(
member_
->
own_local_scope_
)
{
for
(
size_t
i
=
1
;
i
<
member_
->
local_scopes_
.
size
();
++
i
)
{
Scope
*
local_scope
=
member_
->
local_scopes_
[
i
];
if
(
member_
->
global_scope_
->
HasKid
(
local_scope
))
{
member_
->
global_scope_
->
DeleteScope
(
local_scope
);
}
}
}
// member_ must be destructed before gcs_ since the destructor of
// ReferenceCountOpHandle use raw pointers of gcs_ inside.
member_
.
reset
();
...
...
paddle/fluid/inference/CMakeLists.txt
浏览文件 @
44cb70c0
if
(
WITH_TESTING
)
include
(
test.cmake
)
# some generic cmake funtion for inference
include
(
test
s/test
.cmake
)
# some generic cmake funtion for inference
endif
()
# analysis and tensorrt must be added before creating static library,
# otherwise, there would be undefined reference to them in static library.
...
...
paddle/fluid/inference/tests/api/CMakeLists.txt
浏览文件 @
44cb70c0
set
(
INFERENCE_EXTRA_DEPS paddle_inference_api paddle_fluid_api ir_pass_manager analysis_predictor
)
function
(
download_model install_dir model_name
)
if
(
NOT EXISTS
${
install_dir
}
)
inference_download_and_uncompress
(
${
install_dir
}
${
INFERENCE_URL
}
${
model_name
}
)
endif
()
endfunction
()
function
(
download_model_and_data install_dir model_name data_name
)
if
(
NOT EXISTS
${
install_dir
}
)
inference_download_and_uncompress
(
${
install_dir
}
${
INFERENCE_URL
}
${
model_name
}
)
...
...
@@ -13,6 +19,13 @@ function(inference_analysis_api_test target install_dir filename)
ARGS --infer_model=
${
install_dir
}
/model --infer_data=
${
install_dir
}
/data.txt
)
endfunction
()
function
(
inference_analysis_api_test_with_fake_data target install_dir filename model_name
)
download_model
(
${
install_dir
}
${
model_name
}
)
inference_analysis_test
(
${
target
}
SRCS
${
filename
}
EXTRA_DEPS
${
INFERENCE_EXTRA_DEPS
}
ARGS --infer_model=
${
install_dir
}
/model
)
endfunction
()
# RNN1
if
(
NOT APPLE
)
set
(
RNN1_INSTALL_DIR
"
${
INFERENCE_DEMO_INSTALL_DIR
}
/rnn1"
)
...
...
@@ -61,17 +74,13 @@ inference_analysis_api_test(test_analyzer_seq_conv1 ${SEQ_CONV1_INSTALL_DIR} ana
# ocr
set
(
OCR_INSTALL_DIR
"
${
INFERENCE_DEMO_INSTALL_DIR
}
/ocr"
)
if
(
NOT EXISTS
${
OCR_INSTALL_DIR
}
)
inference_download_and_uncompress
(
${
OCR_INSTALL_DIR
}
"http://paddlemodels.cdn.bcebos.com/"
"inference-vis-demos%2Focr.tar.gz"
)
inference_download_and_uncompress
(
${
OCR_INSTALL_DIR
}
"http://paddlemodels.cdn.bcebos.com/"
"inference-vis-demos%2Focr.tar.gz"
)
endif
()
inference_analysis_api_test
(
test_analyzer_ocr
${
OCR_INSTALL_DIR
}
analyzer_vis_tester.cc
)
# resnet50
set
(
RESNET50_INSTALL_DIR
"
${
INFERENCE_DEMO_INSTALL_DIR
}
/resnet50"
)
if
(
NOT EXISTS
${
RESNET50_INSTALL_DIR
}
)
inference_download_and_uncompress
(
${
RESNET50_INSTALL_DIR
}
${
INFERENCE_URL
}
"resnet50_model.tar.gz"
)
endif
()
inference_analysis_test
(
test_analyzer_resnet50 SRCS analyzer_resnet50_tester.cc
EXTRA_DEPS
${
INFERENCE_EXTRA_DEPS
}
ARGS --infer_model=
${
RESNET50_INSTALL_DIR
}
/model
)
inference_analysis_api_test_with_fake_data
(
test_analyzer_resnet50
"
${
INFERENCE_DEMO_INSTALL_DIR
}
/resnet50"
analyzer_resnet50_tester.cc
"resnet50_model.tar.gz"
)
# anakin
if
(
WITH_ANAKIN AND WITH_MKL
)
# only needed in CI
...
...
paddle/fluid/inference/tests/api/analyzer_resnet50_tester.cc
浏览文件 @
44cb70c0
...
...
@@ -30,25 +30,7 @@ void SetConfig(AnalysisConfig *cfg) {
}
void
SetInput
(
std
::
vector
<
std
::
vector
<
PaddleTensor
>>
*
inputs
)
{
PADDLE_ENFORCE_EQ
(
FLAGS_test_all_data
,
0
,
"Only have single batch of data."
);
PaddleTensor
input
;
// channel=3, height/width=318
std
::
vector
<
int
>
shape
({
FLAGS_batch_size
,
3
,
318
,
318
});
input
.
shape
=
shape
;
input
.
dtype
=
PaddleDType
::
FLOAT32
;
// fill input data, for profile easily, do not use random data here.
size_t
size
=
FLAGS_batch_size
*
3
*
318
*
318
;
input
.
data
.
Resize
(
size
*
sizeof
(
float
));
float
*
input_data
=
static_cast
<
float
*>
(
input
.
data
.
data
());
for
(
size_t
i
=
0
;
i
<
size
;
i
++
)
{
*
(
input_data
+
i
)
=
static_cast
<
float
>
(
i
)
/
size
;
}
std
::
vector
<
PaddleTensor
>
input_slots
;
input_slots
.
assign
({
input
});
(
*
inputs
).
emplace_back
(
input_slots
);
SetFakeImageInput
(
inputs
,
FLAGS_infer_model
);
}
// Easy for profiling independently.
...
...
@@ -61,13 +43,6 @@ void profile(bool use_mkldnn = false) {
std
::
vector
<
std
::
vector
<
PaddleTensor
>>
input_slots_all
;
SetInput
(
&
input_slots_all
);
TestPrediction
(
cfg
,
input_slots_all
,
&
outputs
,
FLAGS_num_threads
);
if
(
FLAGS_num_threads
==
1
&&
!
FLAGS_test_all_data
)
{
PADDLE_ENFORCE_EQ
(
outputs
.
size
(),
1UL
);
size_t
size
=
GetSize
(
outputs
[
0
]);
// output is a 512-dimension feature
EXPECT_EQ
(
size
,
512
*
FLAGS_batch_size
);
}
}
TEST
(
Analyzer_resnet50
,
profile
)
{
profile
();
}
...
...
@@ -83,8 +58,7 @@ TEST(Analyzer_resnet50, fuse_statis) {
auto
predictor
=
CreatePaddlePredictor
<
AnalysisConfig
>
(
cfg
);
auto
fuse_statis
=
GetFuseStatis
(
static_cast
<
AnalysisPredictor
*>
(
predictor
.
get
()),
&
num_ops
);
ASSERT_TRUE
(
fuse_statis
.
count
(
"fc_fuse"
));
EXPECT_EQ
(
fuse_statis
.
at
(
"fc_fuse"
),
1
);
LOG
(
INFO
)
<<
"num_ops: "
<<
num_ops
;
}
// Compare result of NativeConfig and AnalysisConfig
...
...
paddle/fluid/inference/tests/api/tester_helper.h
浏览文件 @
44cb70c0
...
...
@@ -25,6 +25,7 @@
#include "paddle/fluid/inference/api/analysis_predictor.h"
#include "paddle/fluid/inference/api/helper.h"
#include "paddle/fluid/inference/api/paddle_inference_pass.h"
#include "paddle/fluid/inference/tests/test_helper.h"
#include "paddle/fluid/platform/profiler.h"
DEFINE_string
(
infer_model
,
""
,
"model path"
);
...
...
@@ -105,6 +106,34 @@ std::unordered_map<std::string, int> GetFuseStatis(PaddlePredictor *predictor,
return
fuse_statis
;
}
void
SetFakeImageInput
(
std
::
vector
<
std
::
vector
<
PaddleTensor
>>
*
inputs
,
const
std
::
string
&
dirname
)
{
// Set fake_image_data
PADDLE_ENFORCE_EQ
(
FLAGS_test_all_data
,
0
,
"Only have single batch of data."
);
std
::
vector
<
std
::
vector
<
int64_t
>>
feed_target_shapes
=
GetFeedTargetShapes
(
dirname
,
true
,
"model"
,
"params"
);
int
dim1
=
feed_target_shapes
[
0
][
1
];
int
dim2
=
feed_target_shapes
[
0
][
2
];
int
dim3
=
feed_target_shapes
[
0
][
3
];
PaddleTensor
input
;
std
::
vector
<
int
>
shape
({
FLAGS_batch_size
,
dim1
,
dim2
,
dim3
});
input
.
shape
=
shape
;
input
.
dtype
=
PaddleDType
::
FLOAT32
;
// fill input data, for profile easily, do not use random data here.
size_t
size
=
FLAGS_batch_size
*
dim1
*
dim2
*
dim3
;
input
.
data
.
Resize
(
size
*
sizeof
(
float
));
float
*
input_data
=
static_cast
<
float
*>
(
input
.
data
.
data
());
for
(
size_t
i
=
0
;
i
<
size
;
i
++
)
{
*
(
input_data
+
i
)
=
static_cast
<
float
>
(
i
)
/
size
;
}
std
::
vector
<
PaddleTensor
>
input_slots
;
input_slots
.
assign
({
input
});
(
*
inputs
).
emplace_back
(
input_slots
);
}
void
TestOneThreadPrediction
(
const
AnalysisConfig
&
config
,
const
std
::
vector
<
std
::
vector
<
PaddleTensor
>>
&
inputs
,
...
...
paddle/fluid/inference/test.cmake
→
paddle/fluid/inference/test
s/test
.cmake
浏览文件 @
44cb70c0
文件已移动
paddle/fluid/inference/tests/test_helper.h
浏览文件 @
44cb70c0
...
...
@@ -18,7 +18,6 @@ limitations under the License. */
#include <string>
#include <vector>
#include "paddle/fluid/framework/ir/graph_to_program_pass.h"
#include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/inference/io.h"
#include "paddle/fluid/platform/profiler.h"
...
...
@@ -94,15 +93,15 @@ void CheckError(const paddle::framework::LoDTensor& output1,
std
::
unique_ptr
<
paddle
::
framework
::
ProgramDesc
>
InitProgram
(
paddle
::
framework
::
Executor
*
executor
,
paddle
::
framework
::
Scope
*
scope
,
const
std
::
string
&
dirname
,
const
bool
is_combined
=
false
)
{
const
std
::
string
&
dirname
,
const
bool
is_combined
=
false
,
const
std
::
string
&
prog_filename
=
"__model_combined__"
,
const
std
::
string
&
param_filename
=
"__params_combined__"
)
{
std
::
unique_ptr
<
paddle
::
framework
::
ProgramDesc
>
inference_program
;
if
(
is_combined
)
{
// All parameters are saved in a single file.
// Hard-coding the file names of program and parameters in unittest.
// The file names should be consistent with that used in Python API
// `fluid.io.save_inference_model`.
std
::
string
prog_filename
=
"__model_combined__"
;
std
::
string
param_filename
=
"__params_combined__"
;
inference_program
=
paddle
::
inference
::
Load
(
executor
,
scope
,
dirname
+
"/"
+
prog_filename
,
dirname
+
"/"
+
param_filename
);
...
...
@@ -115,12 +114,15 @@ std::unique_ptr<paddle::framework::ProgramDesc> InitProgram(
}
std
::
vector
<
std
::
vector
<
int64_t
>>
GetFeedTargetShapes
(
const
std
::
string
&
dirname
,
const
bool
is_combined
=
false
)
{
const
std
::
string
&
dirname
,
const
bool
is_combined
=
false
,
const
std
::
string
&
prog_filename
=
"__model_combined__"
,
const
std
::
string
&
param_filename
=
"__params_combined__"
)
{
auto
place
=
paddle
::
platform
::
CPUPlace
();
auto
executor
=
paddle
::
framework
::
Executor
(
place
);
auto
*
scope
=
new
paddle
::
framework
::
Scope
();
auto
inference_program
=
InitProgram
(
&
executor
,
scope
,
dirname
,
is_combined
);
auto
inference_program
=
InitProgram
(
&
executor
,
scope
,
dirname
,
is_combined
,
prog_filename
,
param_filename
);
auto
&
global_block
=
inference_program
->
Block
(
0
);
const
std
::
vector
<
std
::
string
>&
feed_target_names
=
...
...
@@ -136,15 +138,6 @@ std::vector<std::vector<int64_t>> GetFeedTargetShapes(
return
feed_target_shapes
;
}
void
Compile
(
paddle
::
framework
::
ProgramDesc
*
program
)
{
std
::
unique_ptr
<
paddle
::
framework
::
ir
::
Graph
>
g
(
new
paddle
::
framework
::
ir
::
Graph
(
*
program
));
auto
pass
=
paddle
::
framework
::
ir
::
PassRegistry
::
Instance
().
Get
(
"graph_to_program_pass"
);
pass
->
SetNotOwned
<
paddle
::
framework
::
ProgramDesc
>
(
"program"
,
program
);
pass
->
Apply
(
std
::
move
(
g
));
}
template
<
typename
Place
,
bool
CreateVars
=
true
,
bool
PrepareContext
=
false
>
void
TestInference
(
const
std
::
string
&
dirname
,
const
std
::
vector
<
paddle
::
framework
::
LoDTensor
*>&
cpu_feeds
,
...
...
@@ -182,7 +175,6 @@ void TestInference(const std::string& dirname,
paddle
::
platform
::
DeviceContextPool
::
Instance
().
Get
(
place
));
inference_program
=
InitProgram
(
&
executor
,
scope
,
dirname
,
is_combined
);
}
Compile
(
inference_program
.
get
());
// Disable the profiler and print the timing information
paddle
::
platform
::
DisableProfiler
(
paddle
::
platform
::
EventSortingKey
::
kDefault
,
...
...
@@ -261,5 +253,3 @@ void TestInference(const std::string& dirname,
delete
scope
;
}
USE_PASS
(
graph_to_program_pass
);
paddle/fluid/pybind/pybind.cc
浏览文件 @
44cb70c0
...
...
@@ -742,7 +742,12 @@ All parameter, weight, gradient are variables in Paddle.
will clean up the temp variables at the end of the current iteration.
2. In some NLP model, it may cause the GPU memory is insufficient,
in this case, you should reduce `num_iteration_per_drop_scope`.
)DOC"
);
)DOC"
)
.
def_property
(
"_dry_run"
,
[](
const
ExecutionStrategy
&
self
)
{
return
self
.
dry_run_
;
},
[](
ExecutionStrategy
&
self
,
bool
dry_run
)
{
self
.
dry_run_
=
dry_run
;
});
exec_strategy
.
def_property
(
"use_experimental_executor"
,
...
...
python/paddle/fluid/layers/io.py
浏览文件 @
44cb70c0
...
...
@@ -60,7 +60,7 @@ def data(name,
For example if shape=[1], the resulting shape is [-1, 1].
2. If shape contains -1, such as shape=[1, -1],
append_batch_size will be enforced to be be False (ineffective).
dtype(
int|float
): The type of data : float32, float_16, int etc
dtype(
basestring
): The type of data : float32, float_16, int etc
type(VarType): The output type. By default it is LOD_TENSOR.
lod_level(int): The LoD Level. 0 means the input data is not a sequence.
stop_gradient(bool): A boolean that mentions whether gradient should flow.
...
...
python/paddle/fluid/tests/unittests/test_parallel_executor_dry_run.py
0 → 100644
浏览文件 @
44cb70c0
# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import
paddle.fluid
as
fluid
import
unittest
import
logging
import
six
class
TestBase
(
unittest
.
TestCase
):
def
main
(
self
,
network_func
,
iter
=
100
,
iter_per_pe
=
100
,
use_gpu
=
True
,
use_experimental_executor
=
False
):
if
use_gpu
and
not
fluid
.
core
.
is_compiled_with_cuda
():
logging
.
warning
(
"Paddle is not compiled with CUDA, skip GPU unittests"
)
return
main_prog
=
fluid
.
Program
()
startup_prog
=
fluid
.
Program
()
scope
=
fluid
.
Scope
()
with
fluid
.
program_guard
(
main_prog
,
startup_prog
):
with
fluid
.
scope_guard
(
scope
):
loss
=
network_func
()
fluid
.
Executor
(
fluid
.
CUDAPlace
(
0
)
if
use_gpu
else
fluid
.
CPUPlace
()).
run
(
startup_prog
)
for
_
in
six
.
moves
.
xrange
(
iter
):
exe_strategy
=
fluid
.
ExecutionStrategy
()
exe_strategy
.
_dry_run
=
True
exe_strategy
.
use_experimental_executor
=
use_experimental_executor
pe
=
fluid
.
ParallelExecutor
(
use_cuda
=
True
,
loss_name
=
loss
.
name
,
main_program
=
main_prog
,
exec_strategy
=
exe_strategy
)
for
_
in
six
.
moves
.
xrange
(
iter_per_pe
):
pe
.
run
([])
class
TestMNISTDryRun
(
TestBase
):
def
test_mnist_dry_run
(
self
):
for
use_gpu
in
(
False
,
True
):
for
use_experimental_executor
in
(
False
,
True
):
self
.
main
(
network_func
=
TestMNISTDryRun
.
network_func
,
use_gpu
=
use_gpu
,
use_experimental_executor
=
use_experimental_executor
)
@
staticmethod
def
network_func
():
img
=
fluid
.
layers
.
data
(
name
=
'img'
,
shape
=
[
784
],
dtype
=
'float32'
)
label
=
fluid
.
layers
.
data
(
name
=
'label'
,
shape
=
[
1
],
dtype
=
'int64'
)
hidden
=
img
for
_
in
six
.
moves
.
xrange
(
10
):
hidden
=
fluid
.
layers
.
fc
(
input
=
img
,
size
=
200
,
act
=
'tanh'
)
prediction
=
fluid
.
layers
.
fc
(
input
=
hidden
,
size
=
10
,
act
=
'softmax'
)
loss
=
fluid
.
layers
.
cross_entropy
(
input
=
prediction
,
label
=
label
)
avg_loss
=
fluid
.
layers
.
mean
(
loss
)
fluid
.
optimizer
.
Adam
().
minimize
(
avg_loss
)
return
avg_loss
if
__name__
==
'__main__'
:
unittest
.
main
()
python/paddle/fluid/tests/unittests/test_parallel_executor_mnist.py
浏览文件 @
44cb70c0
...
...
@@ -14,30 +14,18 @@
from
__future__
import
print_function
from
parallel_executor_test_base
import
TestParallelExecutorBase
import
paddle.fluid
as
fluid
import
paddle.fluid.core
as
core
import
numpy
as
np
import
paddle
import
paddle.dataset.mnist
as
mnist
import
unittest
import
os
MNIST_RECORDIO_FILE
=
"./mnist_test_pe.recordio"
import
numpy
as
np
import
paddle.fluid.core
as
core
import
os
import
paddle.fluid
as
fluid
from
parallel_executor_test_base
import
TestParallelExecutorBase
def
simple_fc_net
(
use_feed
):
if
use_feed
:
img
=
fluid
.
layers
.
data
(
name
=
'image'
,
shape
=
[
784
],
dtype
=
'float32'
)
label
=
fluid
.
layers
.
data
(
name
=
'label'
,
shape
=
[
1
],
dtype
=
'int64'
)
else
:
reader
=
fluid
.
layers
.
open_files
(
filenames
=
[
MNIST_RECORDIO_FILE
],
shapes
=
[[
-
1
,
784
],
[
-
1
,
1
]],
lod_levels
=
[
0
,
0
],
dtypes
=
[
'float32'
,
'int64'
])
reader
=
fluid
.
layers
.
io
.
double_buffer
(
reader
)
img
,
label
=
fluid
.
layers
.
read_file
(
reader
)
img
=
fluid
.
layers
.
data
(
name
=
'image'
,
shape
=
[
784
],
dtype
=
'float32'
)
label
=
fluid
.
layers
.
data
(
name
=
'label'
,
shape
=
[
1
],
dtype
=
'int64'
)
hidden
=
img
for
_
in
range
(
4
):
hidden
=
fluid
.
layers
.
fc
(
...
...
@@ -53,17 +41,8 @@ def simple_fc_net(use_feed):
def
fc_with_batchnorm
(
use_feed
):
if
use_feed
:
img
=
fluid
.
layers
.
data
(
name
=
'image'
,
shape
=
[
784
],
dtype
=
'float32'
)
label
=
fluid
.
layers
.
data
(
name
=
'label'
,
shape
=
[
1
],
dtype
=
'int64'
)
else
:
reader
=
fluid
.
layers
.
open_files
(
filenames
=
[
MNIST_RECORDIO_FILE
],
shapes
=
[[
-
1
,
784
],
[
-
1
,
1
]],
lod_levels
=
[
0
,
0
],
dtypes
=
[
'float32'
,
'int64'
])
reader
=
fluid
.
layers
.
io
.
double_buffer
(
reader
)
img
,
label
=
fluid
.
layers
.
read_file
(
reader
)
img
=
fluid
.
layers
.
data
(
name
=
'image'
,
shape
=
[
784
],
dtype
=
'float32'
)
label
=
fluid
.
layers
.
data
(
name
=
'label'
,
shape
=
[
1
],
dtype
=
'int64'
)
hidden
=
img
for
_
in
range
(
1
):
...
...
@@ -88,19 +67,6 @@ class TestMNIST(TestParallelExecutorBase):
@
classmethod
def
setUpClass
(
cls
):
os
.
environ
[
'CPU_NUM'
]
=
str
(
4
)
# Convert mnist to recordio file
with
fluid
.
program_guard
(
fluid
.
Program
(),
fluid
.
Program
()):
reader
=
paddle
.
batch
(
mnist
.
train
(),
batch_size
=
4
)
feeder
=
fluid
.
DataFeeder
(
feed_list
=
[
# order is image and label
fluid
.
layers
.
data
(
name
=
'image'
,
shape
=
[
784
]),
fluid
.
layers
.
data
(
name
=
'label'
,
shape
=
[
1
],
dtype
=
'int64'
),
],
place
=
fluid
.
CPUPlace
())
fluid
.
recordio_writer
.
convert_reader_to_recordio_file
(
MNIST_RECORDIO_FILE
,
reader
,
feeder
)
def
_init_data
(
self
):
np
.
random
.
seed
(
5
)
...
...
@@ -111,10 +77,6 @@ class TestMNIST(TestParallelExecutorBase):
def
_compare_reduce_and_allreduce
(
self
,
model
,
use_cuda
):
if
use_cuda
and
not
core
.
is_compiled_with_cuda
():
return
self
.
check_network_convergence
(
model
,
use_cuda
=
use_cuda
,
use_reduce
=
True
)
self
.
check_network_convergence
(
model
,
use_cuda
=
use_cuda
,
allow_op_delay
=
True
,
use_reduce
=
True
)
img
,
label
=
self
.
_init_data
()
...
...
@@ -140,9 +102,6 @@ class TestMNIST(TestParallelExecutorBase):
def
check_simple_fc_convergence
(
self
,
use_cuda
,
use_reduce
=
False
):
if
use_cuda
and
not
core
.
is_compiled_with_cuda
():
return
self
.
check_network_convergence
(
simple_fc_net
,
use_cuda
=
use_cuda
)
self
.
check_network_convergence
(
simple_fc_net
,
use_cuda
=
use_cuda
,
allow_op_delay
=
True
)
img
,
label
=
self
.
_init_data
()
...
...
@@ -199,8 +158,6 @@ class TestMNIST(TestParallelExecutorBase):
if
use_cuda
and
not
core
.
is_compiled_with_cuda
():
return
self
.
check_network_convergence
(
fc_with_batchnorm
,
use_cuda
=
use_cuda
)
img
,
label
=
self
.
_init_data
()
self
.
check_network_convergence
(
...
...
python/setup.py.in
浏览文件 @
44cb70c0
...
...
@@ -14,7 +14,8 @@ RC = 0
def git_commit():
try:
cmd = ['git', 'rev-parse', 'HEAD']
git_commit = subprocess.Popen(cmd, stdout = subprocess.PIPE).communicate()[0].strip()
git_commit = subprocess.Popen(cmd, stdout = subprocess.PIPE,
cwd="@PADDLE_SOURCE_DIR@").communicate()[0].strip()
except:
git_commit = 'Unknown'
git_commit = git_commit.decode()
...
...
@@ -44,7 +45,7 @@ def get_patch():
def is_taged():
try:
cmd = ['git', 'describe', '--exact-match', '--tags', 'HEAD', '2>/dev/null']
git_tag = subprocess.Popen(cmd, stdout = subprocess.PIPE).communicate()[0].strip()
git_tag = subprocess.Popen(cmd, stdout = subprocess.PIPE
, cwd="@PADDLE_SOURCE_DIR@"
).communicate()[0].strip()
git_tag = git_tag.decode()
except:
return False
...
...
@@ -55,8 +56,7 @@ def is_taged():
return False
def write_version_py(filename='paddle/version.py'):
cnt = '''
# THIS FILE IS GENERATED FROM PADDLEPADDLE SETUP.PY
cnt = '''# THIS FILE IS GENERATED FROM PADDLEPADDLE SETUP.PY
#
full_version = '%(major)d.%(minor)d.%(patch)s'
major = '%(major)d'
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录