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
PaddlePaddle
Paddle
提交
dc847f12
P
Paddle
项目概览
PaddlePaddle
/
Paddle
1 年多 前同步成功
通知
2302
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
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
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
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
提交
dc847f12
编写于
6月 25, 2018
作者:
T
tangwei12
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
bug fix and code optimize
上级
fb7e4791
变更
4
隐藏空白更改
内联
并排
Showing
4 changed file
with
17 addition
and
18 deletion
+17
-18
paddle/fluid/operators/distributed/request_handler_impl.cc
paddle/fluid/operators/distributed/request_handler_impl.cc
+2
-2
paddle/fluid/operators/listen_and_serv_op.cc
paddle/fluid/operators/listen_and_serv_op.cc
+10
-9
paddle/fluid/operators/save_op.cc
paddle/fluid/operators/save_op.cc
+2
-2
python/paddle/fluid/transpiler/distribute_transpiler.py
python/paddle/fluid/transpiler/distribute_transpiler.py
+3
-5
未找到文件。
paddle/fluid/operators/distributed/request_handler_impl.cc
浏览文件 @
dc847f12
...
...
@@ -30,7 +30,7 @@ namespace distributed {
// define LOOKUP_TABLE_PATH for checkpoint notify to save lookup table variables
// to directory specified.
constexpr
char
LOOKUP_TABLE_PATH
[]
=
"
lookup_table_p
ath"
;
constexpr
char
LOOKUP_TABLE_PATH
[]
=
"
kLookupTableP
ath"
;
bool
RequestSendHandler
::
Handle
(
const
std
::
string
&
varname
,
framework
::
Scope
*
scope
,
...
...
@@ -136,7 +136,7 @@ bool RequestCheckpointHandler::Handle(const std::string& varname,
auto
*
lt_var
=
scope
->
FindVar
(
LOOKUP_TABLE_PATH
)
->
GetMutable
<
std
::
string
>
();
lt_var
->
clear
();
lt_var
->
append
(
out_var_name
);
VLOG
(
4
)
<<
"RequestCheckpointHandler update var
lookup_table_p
ath to: "
VLOG
(
4
)
<<
"RequestCheckpointHandler update var
kLookupTableP
ath to: "
<<
out_var_name
;
executor_
->
RunPreparedContext
(
checkpoint_prepared_ctx_
.
get
(),
scope
);
return
true
;
...
...
paddle/fluid/operators/listen_and_serv_op.cc
浏览文件 @
dc847f12
...
...
@@ -206,7 +206,7 @@ void ListenAndServOp::RunAsyncLoop(framework::Executor *executor,
VLOG
(
3
)
<<
"RunAsyncLoop into while"
;
while
(
true
)
{
if
(
rpc_service_
->
IsExit
())
{
LOG
(
INFO
)
<<
"get exit!rpc_processor break!"
;
VLOG
(
4
)
<<
"get exit!rpc_processor break!"
;
break
;
}
...
...
@@ -245,11 +245,11 @@ void ListenAndServOp::RunImpl(const framework::Scope &scope,
PADDLE_ENFORCE
(
!
rpc_service_
);
std
::
string
endpoint
=
Attr
<
std
::
string
>
(
"endpoint"
);
int
checkpoint_
notify_
block_id
=
Attr
<
int
>
(
kCheckpointBlockId
);
int
checkpoint_block_id
=
Attr
<
int
>
(
kCheckpointBlockId
);
LOG
(
INFO
)
<<
"sync_mode:"
<<
sync_mode
<<
", fan_in:"
<<
fan_in
<<
", end_point:"
<<
endpoint
<<
", CheckpointNotify Id: "
<<
checkpoint_notify
_block_id
;
VLOG
(
4
)
<<
"sync_mode:"
<<
sync_mode
<<
", fan_in:"
<<
fan_in
<<
", end_point:"
<<
endpoint
<<
", checkpoint_block_id: "
<<
checkpoint
_block_id
;
rpc_service_
.
reset
(
new
RPCSERVER_T
(
endpoint
,
fan_in
));
...
...
@@ -258,7 +258,7 @@ void ListenAndServOp::RunImpl(const framework::Scope &scope,
request_prefetch_handler_
.
reset
(
new
distributed
::
RequestPrefetchHandler
(
sync_mode
));
request_checkpoint_handler_
.
reset
(
new
distributed
::
RequestCheckpointHandler
(
sync_mode
,
checkpoint_
notify_
block_id
));
sync_mode
,
checkpoint_block_id
));
rpc_service_
->
RegisterRPC
(
distributed
::
kRequestSend
,
request_send_handler_
.
get
());
...
...
@@ -277,8 +277,9 @@ void ListenAndServOp::RunImpl(const framework::Scope &scope,
framework
::
Executor
executor
(
dev_place
);
std
::
shared_ptr
<
framework
::
ExecutorPrepareContext
>
ckpt_pre_context
=
nullptr
;
if
(
checkpoint_notify_block_id
!=
-
1
)
{
auto
ctx
=
executor
.
Prepare
(
*
program
,
checkpoint_notify_block_id
);
if
(
checkpoint_block_id
!=
-
1
)
{
auto
ctx
=
executor
.
Prepare
(
*
program
,
checkpoint_block_id
);
// see: https://stackoverflow.com/a/14856553
ckpt_pre_context
=
std
::
move
(
ctx
);
}
...
...
@@ -335,7 +336,7 @@ void ListenAndServOp::RunImpl(const framework::Scope &scope,
SavePort
();
if
(
sync_mode
)
{
RunSyncLoop
(
&
executor
,
program
,
&
recv_scope
,
prefetch_block_id_list
,
checkpoint_
notify_
block_id
);
checkpoint_block_id
);
}
else
{
RunAsyncLoop
(
&
executor
,
program
);
}
...
...
paddle/fluid/operators/save_op.cc
浏览文件 @
dc847f12
...
...
@@ -31,7 +31,7 @@ namespace operators {
// define LOOKUP_TABLE_PATH for checkpoint notify to save lookup table variables
// to directory specified.
constexpr
char
LOOKUP_TABLE_PATH
[]
=
"
lookup_table_p
ath"
;
constexpr
char
LOOKUP_TABLE_PATH
[]
=
"
kLookupTableP
ath"
;
// TODO(yuyang18): If the functions below are needed by other files, move them
// to paddle::filesystem namespace.
...
...
@@ -138,7 +138,7 @@ class SaveOp : public framework::OperatorBase {
auto
*
lt_var
=
scope
.
FindVar
(
LOOKUP_TABLE_PATH
)
->
GetMutable
<
std
::
string
>
();
PADDLE_ENFORCE
(
lt_var
!=
nullptr
,
"Can not find variable
lookup_table_p
ath for SaveSelectedRows"
);
"Can not find variable
kLookupTableP
ath for SaveSelectedRows"
);
std
::
string
filename
=
lt_var
->
data
();
VLOG
(
4
)
<<
"SaveSelectedRows get File name: "
<<
filename
;
...
...
python/paddle/fluid/transpiler/distribute_transpiler.py
浏览文件 @
dc847f12
...
...
@@ -920,19 +920,17 @@ class DistributeTranspiler(object):
import
os
pserver_program
.
global_block
().
create_var
(
name
=
"
lookup_table_p
ath"
,
name
=
"
kLookupTableP
ath"
,
persistable
=
True
,
type
=
core
.
VarDesc
.
VarType
.
RAW
)
checkpoint_save_block
=
pserver_program
.
create_block
(
pre_block_idx
)
# this 'file_path' do not be used in save lookup table variable
checkpoint_save_block
.
append_op
(
type
=
'save'
,
inputs
=
{
'X'
:
[
self
.
table_name
]},
outputs
=
{},
attrs
=
{
'file_path'
:
"this 'file_path' do not be used in save lookup table variable"
})
attrs
=
{
'file_path'
:
""
})
return
checkpoint_save_block
.
idx
...
...
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