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
csdn_franckjun
Paddle
提交
8ff7df8f
P
Paddle
项目概览
csdn_franckjun
/
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
1
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
1
Issue
1
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
体验新版 GitCode,发现更多精彩内容 >>
未验证
提交
8ff7df8f
编写于
9月 20, 2022
作者:
Y
Yuang Liu
提交者:
GitHub
9月 20, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
dont wait for send op under dygraph pp (#46209)
上级
37ba18bb
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
27 addition
and
30 deletion
+27
-30
python/paddle/distributed/fleet/meta_parallel/pp_utils/p2p_communication.py
...ributed/fleet/meta_parallel/pp_utils/p2p_communication.py
+27
-30
未找到文件。
python/paddle/distributed/fleet/meta_parallel/pp_utils/p2p_communication.py
浏览文件 @
8ff7df8f
...
...
@@ -329,23 +329,21 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
for
d
in
tensor_send_prev
:
if
_in_legacy_dygraph
():
paddle
.
distributed
.
wait
(
d
,
use_calc_stream
=
True
)
tasks
.
append
(
send_partial
(
d
,
dst
=
0
,
nranks
=
mp_degree
,
rank_id
=
mp_rank
,
group
=
_hcg
.
send_prev_group
,
use_calc_stream
=
False
))
else
:
if
_in_legacy_dygraph
():
paddle
.
distributed
.
wait
(
tensor_send_prev
,
use_calc_stream
=
True
)
tasks
.
append
(
send_partial
(
tensor_send_prev
,
send_partial
(
d
,
dst
=
0
,
nranks
=
mp_degree
,
rank_id
=
mp_rank
,
group
=
_hcg
.
send_prev_group
,
use_calc_stream
=
False
))
use_calc_stream
=
False
)
else
:
if
_in_legacy_dygraph
():
paddle
.
distributed
.
wait
(
tensor_send_prev
,
use_calc_stream
=
True
)
send_partial
(
tensor_send_prev
,
dst
=
0
,
nranks
=
mp_degree
,
rank_id
=
mp_rank
,
group
=
_hcg
.
send_prev_group
,
use_calc_stream
=
False
)
if
tensor_recv_prev
is
not
None
:
if
isinstance
(
tensor_recv_prev
,
tuple
):
...
...
@@ -371,23 +369,21 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
for
d
in
tensor_send_next
:
if
_in_legacy_dygraph
():
paddle
.
distributed
.
wait
(
d
,
use_calc_stream
=
True
)
tasks
.
append
(
send_partial
(
d
,
dst
=
1
,
nranks
=
mp_degree
,
rank_id
=
mp_rank
,
group
=
_hcg
.
send_next_group
,
use_calc_stream
=
False
))
else
:
if
_in_legacy_dygraph
():
paddle
.
distributed
.
wait
(
tensor_send_next
,
use_calc_stream
=
True
)
tasks
.
append
(
send_partial
(
tensor_send_next
,
send_partial
(
d
,
dst
=
1
,
nranks
=
mp_degree
,
rank_id
=
mp_rank
,
group
=
_hcg
.
send_next_group
,
use_calc_stream
=
False
))
use_calc_stream
=
False
)
else
:
if
_in_legacy_dygraph
():
paddle
.
distributed
.
wait
(
tensor_send_next
,
use_calc_stream
=
True
)
send_partial
(
tensor_send_next
,
dst
=
1
,
nranks
=
mp_degree
,
rank_id
=
mp_rank
,
group
=
_hcg
.
send_next_group
,
use_calc_stream
=
False
)
if
tensor_recv_next
is
not
None
:
if
isinstance
(
tensor_recv_next
,
tuple
):
...
...
@@ -438,10 +434,11 @@ def _p2p_helper(tensor_send_next, tensor_send_prev, recv_prev, recv_next):
group
=
mp_group
,
use_calc_stream
=
True
))
for
task
in
tasks
:
# wait partial all gather tasks
if
task
is
not
None
:
task
.
wait
()
if
in_dygraph_mode
():
for
task
in
tasks
:
# wait partial all gather tasks
if
task
is
not
None
:
task
.
wait
()
return
tensor_recv_prev
,
tensor_recv_next
...
...
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