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
37bb3342
P
Paddle
项目概览
PaddlePaddle
/
Paddle
大约 1 年 前同步成功
通知
2298
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
体验新版 GitCode,发现更多精彩内容 >>
未验证
提交
37bb3342
编写于
4月 21, 2021
作者:
Y
Yuang Liu
提交者:
GitHub
4月 21, 2021
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
add get_loss_scaling to fleet (#32401)
上级
2b68d20b
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
24 addition
and
15 deletion
+24
-15
python/paddle/distributed/fleet/base/fleet_base.py
python/paddle/distributed/fleet/base/fleet_base.py
+21
-15
python/paddle/fluid/contrib/mixed_precision/decorator.py
python/paddle/fluid/contrib/mixed_precision/decorator.py
+1
-0
python/paddle/fluid/tests/unittests/test_fleet_amp_init.py
python/paddle/fluid/tests/unittests/test_fleet_amp_init.py
+2
-0
未找到文件。
python/paddle/distributed/fleet/base/fleet_base.py
浏览文件 @
37bb3342
...
...
@@ -1041,6 +1041,26 @@ class Fleet(object):
# imitate target optimizer retrieval
return
self
.
user_defined_optimizer
.
clear_grad
()
def
_get_amp_optimizer
(
self
):
# imitate target optimizer retrieval
amp_optimizer
=
None
for
optimizer
in
self
.
strategy_compiler
.
_get_applied_meta_optimizer
():
if
hasattr
(
optimizer
,
'amp_init'
):
amp_optimizer
=
optimizer
break
if
amp_optimizer
is
None
:
if
hasattr
(
self
.
user_defined_optimizer
,
'amp_init'
):
amp_optimizer
=
self
.
user_defined_optimizer
assert
amp_optimizer
is
not
None
,
\
"amp_init can only be used when the amp(auto mixed precision) strategy is turned on."
return
amp_optimizer
def
get_loss_scaling
(
self
):
amp_optimizer
=
self
.
_get_amp_optimizer
()
return
amp_optimizer
.
get_loss_scaling
()
def
amp_init
(
self
,
place
,
scope
=
None
,
...
...
@@ -1101,21 +1121,7 @@ class Fleet(object):
if paddle.is_compiled_with_cuda() and len(paddle.static.cuda_places()) > 0:
run_example_code()
"""
# imitate target optimizer retrieval
amp_optimizer
=
None
for
optimizer
in
self
.
strategy_compiler
.
_get_applied_meta_optimizer
():
if
hasattr
(
optimizer
,
'amp_init'
):
amp_optimizer
=
optimizer
break
if
amp_optimizer
is
None
:
if
hasattr
(
self
.
user_defined_optimizer
,
'amp_init'
):
amp_optimizer
=
self
.
user_defined_optimizer
assert
amp_optimizer
is
not
None
,
\
"amp_init can only be used when the amp(auto mixed precision) strategy is turned on."
amp_optimizer
=
self
.
_get_amp_optimizer
()
return
amp_optimizer
.
amp_init
(
place
,
scope
,
test_program
,
use_fp16_test
)
def
_final_strategy
(
self
):
...
...
python/paddle/fluid/contrib/mixed_precision/decorator.py
浏览文件 @
37bb3342
...
...
@@ -98,6 +98,7 @@ class OptimizerWithMixedPrecision(object):
def
get_loss_scaling
(
self
):
"""Return the real-time loss scaling factor.
"""
assert
self
.
_loss_scaling
is
not
None
,
'Call minimize() before calling get_loss_scaling()'
return
self
.
_loss_scaling
def
get_scaled_loss
(
self
):
...
...
python/paddle/fluid/tests/unittests/test_fleet_amp_init.py
浏览文件 @
37bb3342
...
...
@@ -70,6 +70,8 @@ class TestFleetAMPInit(unittest.TestCase):
optimizer
=
fleet
.
distributed_optimizer
(
optimizer
)
optimizer
.
minimize
(
cost
)
loss_scale
=
optimizer
.
get_loss_scaling
()
place
=
paddle
.
CUDAPlace
(
0
)
exe
=
paddle
.
static
.
Executor
(
place
)
...
...
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