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
PaddleRec
提交
9438cb36
P
PaddleRec
项目概览
PaddlePaddle
/
PaddleRec
通知
68
Star
12
Fork
5
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
27
列表
看板
标记
里程碑
合并请求
10
Wiki
1
Wiki
分析
仓库
DevOps
项目成员
Pages
P
PaddleRec
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
27
Issue
27
列表
看板
标记
里程碑
合并请求
10
合并请求
10
Pages
分析
分析
仓库分析
DevOps
Wiki
1
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
9438cb36
编写于
5月 13, 2020
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
上级
77a2da6d
变更
5
显示空白变更内容
内联
并排
Showing
5 changed file
with
28 addition
and
14 deletion
+28
-14
fleet_rec/core/trainers/cluster_trainer.py
fleet_rec/core/trainers/cluster_trainer.py
+3
-1
fleet_rec/core/trainers/transpiler_trainer.py
fleet_rec/core/trainers/transpiler_trainer.py
+2
-1
fleet_rec/run.py
fleet_rec/run.py
+4
-4
models/rank/dnn/submit.sh
models/rank/dnn/submit.sh
+2
-2
models/rank/dnn/worker.sh
models/rank/dnn/worker.sh
+17
-6
未找到文件。
fleet_rec/core/trainers/cluster_trainer.py
浏览文件 @
9438cb36
...
...
@@ -23,6 +23,7 @@ import paddle.fluid as fluid
from
paddle.fluid.incubate.fleet.parameter_server.distribute_transpiler
import
fleet
from
paddle.fluid.incubate.fleet.parameter_server.distribute_transpiler.distributed_strategy
import
StrategyFactory
from
paddle.fluid.incubate.fleet.base.role_maker
import
PaddleCloudRoleMaker
from
paddle.fluid.incubate.fleet.base.role_maker
import
MPISymetricRoleMaker
from
fleetrec.core.utils
import
envs
from
fleetrec.core.trainers.transpiler_trainer
import
TranspileTrainer
...
...
@@ -30,7 +31,8 @@ from fleetrec.core.trainers.transpiler_trainer import TranspileTrainer
class
ClusterTrainer
(
TranspileTrainer
):
def
processor_register
(
self
):
role
=
PaddleCloudRoleMaker
()
#role = PaddleCloudRoleMaker()
role
=
MPISymetricRoleMaker
()
fleet
.
init
(
role
)
if
fleet
.
is_server
():
...
...
fleet_rec/core/trainers/transpiler_trainer.py
浏览文件 @
9438cb36
...
...
@@ -72,7 +72,8 @@ class TranspileTrainer(Trainer):
train_data_path
=
envs
.
get_global_env
(
"test_data_path"
,
None
,
namespace
)
threads
=
int
(
envs
.
get_runtime_environ
(
"train.trainer.threads"
))
#threads = int(envs.get_runtime_environ("train.trainer.threads"))
threads
=
2
batch_size
=
envs
.
get_global_env
(
"batch_size"
,
None
,
namespace
)
reader_class
=
envs
.
get_global_env
(
"class"
,
None
,
namespace
)
abs_dir
=
os
.
path
.
dirname
(
os
.
path
.
abspath
(
__file__
))
...
...
fleet_rec/run.py
浏览文件 @
9438cb36
...
...
@@ -110,7 +110,6 @@ def single_engine(args):
def
cluster_engine
(
args
):
from
fleetrec.core.engine.cluster.cluster
import
ClusterEngine
def
update_workspace
(
cluster_envs
):
workspace
=
cluster_envs
.
get
(
"engine_workspace"
,
None
)
...
...
@@ -131,6 +130,7 @@ def cluster_engine(args):
cluster_envs
[
name
]
=
value
def
master
():
from
fleetrec.core.engine.cluster.cluster
import
ClusterEngine
with
open
(
args
.
backend
,
'r'
)
as
rb
:
_envs
=
yaml
.
load
(
rb
.
read
(),
Loader
=
yaml
.
FullLoader
)
...
...
@@ -155,10 +155,10 @@ def cluster_engine(args):
print
(
"launch {} engine with cluster to with model: {}"
.
format
(
trainer
,
args
.
model
))
set_runtime_envs
(
cluster_envs
,
args
.
model
)
launch
=
LocalClusterEngine
(
cluster_envs
,
args
.
model
)
return
launch
trainer
=
TrainerFactory
.
create
(
args
.
model
)
return
trainer
if
args
.
role
==
"
worker
"
:
if
args
.
role
==
"
WORKER
"
:
return
worker
()
else
:
return
master
()
...
...
models/rank/dnn/submit.sh
浏览文件 @
9438cb36
...
...
@@ -29,8 +29,8 @@ function package() {
cp
${
engine_submit_qconf
}
${
temp
}
echo
"copy job.sh from "
${
engine_worker
}
" to "
${
temp
}
mkdir
-p
${
temp
}
/package
/python
cp
-r
${
engine_package_python
}
/
*
${
temp
}
/package/python
/
mkdir
-p
${
temp
}
/package
cp
-r
${
engine_package_python
}
${
temp
}
/package
/
echo
"copy python from "
${
engine_package_python
}
" to "
${
temp
}
mkdir
${
temp
}
/package/whl
...
...
models/rank/dnn/worker.sh
浏览文件 @
9438cb36
...
...
@@ -16,10 +16,10 @@ declare g_run_stage=""
# ---------------------------------------------------------------------------- #
# const define #
# ---------------------------------------------------------------------------- #
declare
-r
FLAGS_communicator_thread_pool_size
=
5
declare
-r
FLAGS_communicator_send_queue_size
=
18
declare
-r
FLAGS_communicator_thread_pool_size
=
20
declare
-r
FLAGS_communicator_max_merge_var_num
=
18
export
FLAGS_communicator_thread_pool_size
=
5
export
FLAGS_communicator_send_queue_size
=
18
export
FLAGS_communicator_thread_pool_size
=
20
export
FLAGS_communicator_max_merge_var_num
=
18
################################################################################
#-----------------------------------------------------------------------------------------------------------------
...
...
@@ -44,9 +44,20 @@ function env_prepare() {
WORKDIR
=
$(
pwd
)
mpirun
-npernode
1
mv
package/
*
./
echo
"current:"
$WORKDIR
export
LIBRARY_PATH
=
$WORKDIR
/python/lib:
$LIBRARY_PATH
mpirun
-npernode
1 python/bin/python
-m
pip
install
whl/fleet_rec-0.0.2-py2-none-any.whl
--index-url
=
http://pip.baidu.com/pypi/simple
--trusted-host
pip.baidu.com
>
/dev/null
mpirun
-npernode
1
tar
-zxvf
python.tar.gz
>
/dev/null
export
PYTHONPATH
=
$WORKDIR
/python/
export
PYTHONROOT
=
$WORKDIR
/python/
export
LIBRARY_PATH
=
$PYTHONPATH
/lib:
$LIBRARY_PATH
export
LD_LIBRARY_PATH
=
$PYTHONPATH
/lib:
$LD_LIBRARY_PATH
export
PATH
=
$PYTHONPATH
/bin:
$PATH
export
LIBRARY_PATH
=
$PYTHONROOT
/lib:
$LIBRARY_PATH
python
-c
"print('heheda')"
mpirun
-npernode
1 python/bin/python
-m
pip uninstall
-y
fleet-rec
mpirun
-npernode
1 python/bin/python
-m
pip
install
whl/fleet_rec-0.0.2-py2-none-any.whl
--index-url
=
http://pip.baidu.com/pypi/simple
--trusted-host
pip.baidu.com
check_error
}
...
...
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