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
e7986cb4
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
e7986cb4
编写于
4月 20, 2020
作者:
T
tangwei
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
fix readme
上级
3ccb3ea7
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
17 addition
and
30 deletion
+17
-30
fleetrec/core/factory.py
fleetrec/core/factory.py
+1
-1
fleetrec/run.py
fleetrec/run.py
+16
-29
未找到文件。
fleetrec/core/factory.py
浏览文件 @
e7986cb4
...
...
@@ -46,7 +46,7 @@ class TrainerFactory(object):
if
trainer_abs
is
None
:
if
not
os
.
path
.
exists
(
train_mode
)
or
os
.
path
.
isfile
(
train_mode
):
raise
ValueError
(
"trainer {} can not be recognized"
)
raise
ValueError
(
"trainer {} can not be recognized"
.
format
(
train_mode
)
)
trainer_abs
=
train_mode
train_mode
=
"UserDefineTrainer"
...
...
fleetrec/run.py
浏览文件 @
e7986cb4
...
...
@@ -16,6 +16,8 @@ def set_runtime_envs(cluster_envs, engine_yaml):
if
engine_yaml
is
not
None
:
with
open
(
engine_yaml
,
'r'
)
as
rb
:
_envs
=
yaml
.
load
(
rb
.
read
(),
Loader
=
yaml
.
FullLoader
)
else
:
_envs
=
{}
if
cluster_envs
is
None
:
cluster_envs
=
{}
...
...
@@ -24,15 +26,6 @@ def set_runtime_envs(cluster_envs, engine_yaml):
print
(
envs
.
pretty_print_envs
(
cluster_envs
,
(
"Runtime Envs"
,
"Value"
)))
def
engine_registry
():
engines
[
"TRAINSPILER"
][
"SINGLE"
]
=
single_engine
engines
[
"TRAINSPILER"
][
"LOCAL_CLUSTER"
]
=
local_cluster_engine
engines
[
"TRAINSPILER"
][
"CLUSTER"
]
=
cluster_engine
engines
[
"PSLIB"
][
"SINGLE"
]
=
local_mpi_engine
engines
[
"PSLIB"
][
"LOCAL_CLUSTER"
]
=
local_mpi_engine
engines
[
"PSLIB"
][
"CLUSTER"
]
=
cluster_mpi_engine
def
get_engine
(
engine
):
engine
=
engine
.
upper
()
if
version
.
is_transpiler
():
...
...
@@ -47,7 +40,7 @@ def get_engine(engine):
def
single_engine
(
args
):
print
(
"use single engine to run model: {}"
.
format
(
args
.
model
))
single_envs
=
{
"trainer.trainer"
:
"SingleTrain
ing
"
}
single_envs
=
{
"trainer.trainer"
:
"SingleTrain
er
"
}
set_runtime_envs
(
single_envs
,
args
.
engine_extras
)
...
...
@@ -58,7 +51,7 @@ def single_engine(args):
def
cluster_engine
(
args
):
print
(
"launch cluster engine with cluster to run model: {}"
.
format
(
args
.
model
))
cluster_envs
=
{
"trainer.trainer"
:
"ClusterTrain
ing
"
}
cluster_envs
=
{
"trainer.trainer"
:
"ClusterTrain
er
"
}
set_runtime_envs
(
cluster_envs
,
args
.
engine_extras
)
envs
.
set_runtime_envions
(
cluster_envs
)
...
...
@@ -69,7 +62,7 @@ def cluster_engine(args):
def
cluster_mpi_engine
(
args
):
print
(
"launch cluster engine with cluster to run model: {}"
.
format
(
args
.
model
))
cluster_envs
=
{
"trainer.trainer"
:
"Ctr
Training
"
}
cluster_envs
=
{
"trainer.trainer"
:
"Ctr
CodingTrainer
"
}
set_runtime_envs
(
cluster_envs
,
args
.
engine_extras
)
trainer
=
TrainerFactory
.
create
(
args
.
model
)
...
...
@@ -85,7 +78,7 @@ def local_cluster_engine(args):
cluster_envs
[
"worker_num"
]
=
1
cluster_envs
[
"start_port"
]
=
36001
cluster_envs
[
"log_dir"
]
=
"logs"
cluster_envs
[
"trainer.trainer"
]
=
"ClusterTrain
ing
"
cluster_envs
[
"trainer.trainer"
]
=
"ClusterTrain
er
"
cluster_envs
[
"trainer.strategy.mode"
]
=
"async"
set_runtime_envs
(
cluster_envs
,
args
.
engine_extras
)
...
...
@@ -104,28 +97,22 @@ def local_mpi_engine(args):
if
not
mpi
:
raise
RuntimeError
(
"can not find mpirun, please check environment"
)
cluster_envs
=
{
"mpirun"
:
mpi
,
"trainer.trainer"
:
"Ctr
Training
"
,
"log_dir"
:
"logs"
}
cluster_envs
=
{
"mpirun"
:
mpi
,
"trainer.trainer"
:
"Ctr
CodingTrainer
"
,
"log_dir"
:
"logs"
}
set_runtime_envs
(
cluster_envs
,
args
.
engine_extras
)
launch
=
LocalMPIEngine
(
cluster_envs
,
args
.
model
)
return
launch
#
# def yaml_engine(engine_yaml, model_yaml):
# with open(engine_yaml, 'r') as rb:
# _config = yaml.load(rb.read(), Loader=yaml.FullLoader)
# assert _config is not None
#
# envs.set_global_envs(_config)
#
# train_location = envs.get_global_env("engine.file")
# train_dirname = os.path.dirname(train_location)
# base_name = os.path.splitext(os.path.basename(train_location))[0]
# sys.path.append(train_dirname)
# trainer_class = envs.lazy_instance(base_name, "UserDefineTraining")
# trainer = trainer_class(model_yaml)
# return trainer
def
engine_registry
():
engines
[
"TRAINSPILER"
][
"SINGLE"
]
=
single_engine
engines
[
"TRAINSPILER"
][
"LOCAL_CLUSTER"
]
=
local_cluster_engine
engines
[
"TRAINSPILER"
][
"CLUSTER"
]
=
cluster_engine
engines
[
"PSLIB"
][
"SINGLE"
]
=
local_mpi_engine
engines
[
"PSLIB"
][
"LOCAL_CLUSTER"
]
=
local_mpi_engine
engines
[
"PSLIB"
][
"CLUSTER"
]
=
cluster_mpi_engine
engine_registry
()
if
__name__
==
"__main__"
:
parser
=
argparse
.
ArgumentParser
(
description
=
'fleet-rec run'
)
...
...
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