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
马赛克医院
Paddle
提交
8c516a24
P
Paddle
项目概览
马赛克医院
/
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
体验新版 GitCode,发现更多精彩内容 >>
提交
8c516a24
编写于
1月 15, 2019
作者:
Q
Qiao Longfei
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
remote min_row_size_to_use_multithread in adam interface test=develop
上级
7fd15ce5
变更
4
隐藏空白更改
内联
并排
Showing
4 changed file
with
9 addition
and
15 deletion
+9
-15
paddle/fluid/API.spec
paddle/fluid/API.spec
+1
-1
paddle/fluid/operators/optimizers/adam_op.cc
paddle/fluid/operators/optimizers/adam_op.cc
+1
-1
paddle/fluid/operators/optimizers/adam_op.h
paddle/fluid/operators/optimizers/adam_op.h
+5
-5
python/paddle/fluid/optimizer.py
python/paddle/fluid/optimizer.py
+2
-8
未找到文件。
paddle/fluid/API.spec
浏览文件 @
8c516a24
...
...
@@ -418,7 +418,7 @@ paddle.fluid.optimizer.AdagradOptimizer.__init__ ArgSpec(args=['self', 'learning
paddle.fluid.optimizer.AdagradOptimizer.apply_gradients ArgSpec(args=['self', 'params_grads'], varargs=None, keywords=None, defaults=None)
paddle.fluid.optimizer.AdagradOptimizer.backward ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set', 'callbacks'], varargs=None, keywords=None, defaults=(None, None, None, None))
paddle.fluid.optimizer.AdagradOptimizer.minimize ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None, None))
paddle.fluid.optimizer.AdamOptimizer.__init__ ArgSpec(args=['self', 'learning_rate', 'beta1', 'beta2', 'epsilon', 'regularization', 'name', 'lazy_mode'
, 'min_row_size_to_use_multithread'], varargs=None, keywords=None, defaults=(0.001, 0.9, 0.999, 1e-08, None, None, False, 0
))
paddle.fluid.optimizer.AdamOptimizer.__init__ ArgSpec(args=['self', 'learning_rate', 'beta1', 'beta2', 'epsilon', 'regularization', 'name', 'lazy_mode'
], varargs=None, keywords=None, defaults=(0.001, 0.9, 0.999, 1e-08, None, None, False
))
paddle.fluid.optimizer.AdamOptimizer.apply_gradients ArgSpec(args=['self', 'params_grads'], varargs=None, keywords=None, defaults=None)
paddle.fluid.optimizer.AdamOptimizer.backward ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set', 'callbacks'], varargs=None, keywords=None, defaults=(None, None, None, None))
paddle.fluid.optimizer.AdamOptimizer.minimize ArgSpec(args=['self', 'loss', 'startup_program', 'parameter_list', 'no_grad_set'], varargs=None, keywords=None, defaults=(None, None, None))
...
...
paddle/fluid/operators/optimizers/adam_op.cc
浏览文件 @
8c516a24
...
...
@@ -120,7 +120,7 @@ class AdamOpMaker : public framework::OpProtoAndCheckerMaker {
"min_row_size_to_use_multithread and "
"inner_op_parallelism is larger then 0, sparse update "
"will run in multithread mode"
)
.
SetDefault
(
0
);
.
SetDefault
(
100
0
);
AddComment
(
R"DOC(
Adam Optimizer.
...
...
paddle/fluid/operators/optimizers/adam_op.h
浏览文件 @
8c516a24
...
...
@@ -494,16 +494,16 @@ class AdamOpKernel : public framework::OpKernel<T> {
<<
" min_row_size_to_use_multithread="
<<
min_row_size_to_use_multithread
;
if
(
FLAGS_inner_op_parallelism
>
10
)
{
LOG
(
WARNING
)
<<
"FLAGS_inner_op_parallelism "
<<
FLAGS_inner_op_parallelism
<<
" is two large!"
;
VLOG
(
1
)
<<
"FLAGS_inner_op_parallelism "
<<
FLAGS_inner_op_parallelism
<<
" is two large!"
;
}
auto
&
grad_rows
=
grad_merge
.
rows
();
std
::
unordered_map
<
size_t
,
int
>
row_id_to_grad_row_offset
;
size_t
param_row_count
=
param
.
numel
()
/
row_numel
;
if
(
param_row_count
<
1000
)
{
LOG
(
WARNING
)
<<
"param_row_count should be larger then 1000 to use "
"multi thread, currently "
<<
param_row_count
;
VLOG
(
1
)
<<
"param_row_count should be larger then 1000 to use "
"multi thread, currently "
<<
param_row_count
;
}
for
(
size_t
i
=
0
;
i
<
grad_rows
.
size
();
++
i
)
{
row_id_to_grad_row_offset
[
grad_rows
[
i
]]
=
i
;
...
...
python/paddle/fluid/optimizer.py
浏览文件 @
8c516a24
...
...
@@ -734,8 +734,6 @@ class AdamOptimizer(Optimizer):
may be very slow. The lazy mode only update the element that has gradient is the current
mini-batch, so it will be much more faster. But this mode has different semantics with the
original Adam algorithm and may lead to different result.
min_row_size_to_use_multithread: if adam use sparse update and the param rows is very large,
you can use FLAGS_inner_op_parallelism and this flag to enable multi thread optimize.
Examples:
.. code-block:: python
...
...
@@ -756,8 +754,7 @@ class AdamOptimizer(Optimizer):
epsilon
=
1e-8
,
regularization
=
None
,
name
=
None
,
lazy_mode
=
False
,
min_row_size_to_use_multithread
=
0
):
lazy_mode
=
False
):
assert
learning_rate
is
not
None
assert
beta1
is
not
None
assert
beta2
is
not
None
...
...
@@ -771,7 +768,6 @@ class AdamOptimizer(Optimizer):
self
.
_beta2
=
beta2
self
.
_epsilon
=
epsilon
self
.
_lazy_mode
=
lazy_mode
self
.
_min_row_size_to_use_multithread
=
min_row_size_to_use_multithread
def
_create_accumulators
(
self
,
block
,
parameters
):
assert
isinstance
(
block
,
framework
.
Block
)
...
...
@@ -826,9 +822,7 @@ class AdamOptimizer(Optimizer):
"beta1"
:
self
.
_beta1
,
"beta2"
:
self
.
_beta2
,
"epsilon"
:
self
.
_epsilon
,
"lazy_mode"
:
self
.
_lazy_mode
,
"min_row_size_to_use_multithread"
:
self
.
_min_row_size_to_use_multithread
"lazy_mode"
:
self
.
_lazy_mode
},
stop_gradient
=
True
)
...
...
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