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
BaiXuePrincess
Paddle
提交
8126a41d
P
Paddle
项目概览
BaiXuePrincess
/
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
0
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
0
Issue
0
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
未验证
提交
8126a41d
编写于
1月 20, 2021
作者:
L
lilong12
提交者:
GitHub
1月 20, 2021
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
fix the bug of all_reduce pipeline gradient multiple times (#30437)
* update, test=develop
上级
621bc4f7
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
8 addition
and
5 deletion
+8
-5
python/paddle/distributed/fleet/meta_optimizers/pipeline_optimizer.py
...e/distributed/fleet/meta_optimizers/pipeline_optimizer.py
+4
-0
python/paddle/fluid/tests/unittests/CMakeLists.txt
python/paddle/fluid/tests/unittests/CMakeLists.txt
+4
-5
未找到文件。
python/paddle/distributed/fleet/meta_optimizers/pipeline_optimizer.py
浏览文件 @
8126a41d
...
@@ -233,6 +233,7 @@ class PipelineOptimizer(MetaOptimizerBase):
...
@@ -233,6 +233,7 @@ class PipelineOptimizer(MetaOptimizerBase):
block
=
self
.
main_program_list
[
ring_id
-
1
][
'program'
].
global_block
()
block
=
self
.
main_program_list
[
ring_id
-
1
][
'program'
].
global_block
()
origin_block
=
self
.
main_program
.
global_block
()
origin_block
=
self
.
main_program
.
global_block
()
grad
=
None
grad
=
None
processed_param_name
=
set
()
for
idx
,
op
in
reversed
(
list
(
enumerate
(
block
.
ops
))):
for
idx
,
op
in
reversed
(
list
(
enumerate
(
block
.
ops
))):
if
is_backward_op
(
op
)
and
\
if
is_backward_op
(
op
)
and
\
OP_ROLE_VAR_KEY
in
op
.
attr_names
:
OP_ROLE_VAR_KEY
in
op
.
attr_names
:
...
@@ -242,7 +243,10 @@ class PipelineOptimizer(MetaOptimizerBase):
...
@@ -242,7 +243,10 @@ class PipelineOptimizer(MetaOptimizerBase):
assert
len
(
op_role_var
)
%
2
==
0
assert
len
(
op_role_var
)
%
2
==
0
offset
=
idx
offset
=
idx
for
i
in
range
(
0
,
len
(
op_role_var
),
2
):
for
i
in
range
(
0
,
len
(
op_role_var
),
2
):
param_name
=
op_role_var
[
i
]
param
=
block
.
vars
[
op_role_var
[
i
]]
param
=
block
.
vars
[
op_role_var
[
i
]]
if
param_name
in
processed_param_name
:
continue
processed_param_name
.
add
(
param_name
)
grad
=
block
.
vars
[
op_role_var
[
i
+
1
]]
grad
=
block
.
vars
[
op_role_var
[
i
+
1
]]
origin_param
=
origin_block
.
vars
[
op_role_var
[
i
]]
origin_param
=
origin_block
.
vars
[
op_role_var
[
i
]]
if
origin_param
.
is_distributed
:
if
origin_param
.
is_distributed
:
...
...
python/paddle/fluid/tests/unittests/CMakeLists.txt
浏览文件 @
8126a41d
...
@@ -10,7 +10,7 @@ if(NOT WITH_NCCL)
...
@@ -10,7 +10,7 @@ if(NOT WITH_NCCL)
endif
()
endif
()
string
(
REPLACE
".py"
""
DIST_TEST_OPS
"
${
DIST_TEST_OPS
}
"
)
string
(
REPLACE
".py"
""
DIST_TEST_OPS
"
${
DIST_TEST_OPS
}
"
)
list
(
APPEND DIST_TEST_OPS test_parallel_dygraph_mnist
)
list
(
APPEND DIST_TEST_OPS test_parallel_dygraph_mnist
)
#
list(APPEND DIST_TEST_OPS test_pipeline)
list
(
APPEND DIST_TEST_OPS test_pipeline
)
list
(
APPEND DIST_TEST_OPS test_parallel_dygraph_se_resnext
)
list
(
APPEND DIST_TEST_OPS test_parallel_dygraph_se_resnext
)
list
(
APPEND DIST_TEST_OPS test_parallel_dygraph_sparse_embedding
)
list
(
APPEND DIST_TEST_OPS test_parallel_dygraph_sparse_embedding
)
list
(
APPEND DIST_TEST_OPS test_parallel_dygraph_sparse_embedding_over_height
)
list
(
APPEND DIST_TEST_OPS test_parallel_dygraph_sparse_embedding_over_height
)
...
@@ -62,7 +62,6 @@ list(APPEND MIXED_DIST_TEST_OPS test_fleet_auto)
...
@@ -62,7 +62,6 @@ list(APPEND MIXED_DIST_TEST_OPS test_fleet_auto)
foreach
(
TEST_OP
${
MIXED_DIST_TEST_OPS
}
)
foreach
(
TEST_OP
${
MIXED_DIST_TEST_OPS
}
)
list
(
REMOVE_ITEM TEST_OPS
${
TEST_OP
}
)
list
(
REMOVE_ITEM TEST_OPS
${
TEST_OP
}
)
endforeach
()
endforeach
()
list
(
REMOVE_ITEM TEST_OPS test_pipeline
)
if
(
NOT WITH_GPU OR WIN32
)
if
(
NOT WITH_GPU OR WIN32
)
LIST
(
REMOVE_ITEM TEST_OPS test_c_comm_init_all_op
)
LIST
(
REMOVE_ITEM TEST_OPS test_c_comm_init_all_op
)
...
@@ -826,9 +825,9 @@ if(WITH_GPU AND NOT WIN32)
...
@@ -826,9 +825,9 @@ if(WITH_GPU AND NOT WIN32)
set_tests_properties
(
test_collective_allgather_api PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allgather_api PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_broadcast_api PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_broadcast_api PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allreduce_api PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_collective_allreduce_api PROPERTIES TIMEOUT 120
)
#
if(WITH_DISTRIBUTE)
if
(
WITH_DISTRIBUTE
)
#
set_tests_properties(test_pipeline PROPERTIES TIMEOUT 120)
set_tests_properties
(
test_pipeline PROPERTIES TIMEOUT 120
)
#
endif()
endif
()
set_tests_properties
(
test_reducescatter_api PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_reducescatter_api PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_broadcast PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_broadcast PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_reducescatter PROPERTIES TIMEOUT 120
)
set_tests_properties
(
test_reducescatter PROPERTIES TIMEOUT 120
)
...
...
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