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
264e76ca
P
Paddle
项目概览
PaddlePaddle
/
Paddle
7 个月 前同步成功
通知
2274
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
前往新版Gitcode,体验更适合开发者的 AI 搜索 >>
未验证
提交
264e76ca
编写于
9月 09, 2020
作者:
L
LielinJiang
提交者:
GitHub
9月 09, 2020
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
[cherrypick-beta-2.0]Fix kl_div,conv and summary api bug (#27195)
* fix some bug
上级
ed52b009
变更
8
隐藏空白更改
内联
并排
Showing
8 changed file
with
115 addition
and
32 deletion
+115
-32
paddle/fluid/operators/kldiv_loss_op.h
paddle/fluid/operators/kldiv_loss_op.h
+5
-1
python/paddle/fluid/tests/unittests/test_kldiv_loss_op.py
python/paddle/fluid/tests/unittests/test_kldiv_loss_op.py
+7
-1
python/paddle/hapi/model.py
python/paddle/hapi/model.py
+7
-2
python/paddle/hapi/model_summary.py
python/paddle/hapi/model_summary.py
+61
-18
python/paddle/nn/functional/loss.py
python/paddle/nn/functional/loss.py
+2
-2
python/paddle/nn/layer/conv.py
python/paddle/nn/layer/conv.py
+1
-1
python/paddle/nn/layer/loss.py
python/paddle/nn/layer/loss.py
+10
-7
python/paddle/tests/test_model.py
python/paddle/tests/test_model.py
+22
-0
未找到文件。
paddle/fluid/operators/kldiv_loss_op.h
浏览文件 @
264e76ca
...
...
@@ -72,7 +72,11 @@ class KLDivLossKernel : public framework::OpKernel<T> {
loss_t
.
device
(
place
)
=
output
;
}
else
if
(
"batchmean"
==
reduction
)
{
auto
output_sum
=
output
.
sum
();
loss_t
.
device
(
place
)
=
output_sum
/
output_sum
.
constant
(
n
);
if
(
n
>
0
)
{
loss_t
.
device
(
place
)
=
output_sum
/
output_sum
.
constant
(
n
);
}
else
{
loss_t
.
device
(
place
)
=
output_sum
;
}
}
else
if
(
"mean"
==
reduction
)
{
loss_t
.
device
(
place
)
=
output
.
mean
();
}
else
if
(
"sum"
==
reduction
)
{
...
...
python/paddle/fluid/tests/unittests/test_kldiv_loss_op.py
浏览文件 @
264e76ca
...
...
@@ -24,7 +24,10 @@ def kldiv_loss(x, target, reduction):
loss
=
np
.
where
(
target
>=
0
,
output
,
np
.
zeros_like
(
x
))
if
reduction
==
"batchmean"
:
return
loss
.
sum
()
/
x
.
shape
[
0
]
if
len
(
x
.
shape
)
>
0
:
return
loss
.
sum
()
/
x
.
shape
[
0
]
else
:
return
loss
.
sum
()
if
reduction
==
"mean"
:
return
loss
.
mean
()
if
reduction
==
"sum"
:
...
...
@@ -93,6 +96,9 @@ class TestKLDivLossDygraph(unittest.TestCase):
def
test_kl_loss_batchmean
(
self
):
self
.
run_kl_loss
(
'batchmean'
)
def
test_kl_loss_batchmean_shape
(
self
):
self
.
run_kl_loss
(
'batchmean'
,
())
def
test_kl_loss_mean
(
self
):
self
.
run_kl_loss
(
'mean'
)
...
...
python/paddle/hapi/model.py
浏览文件 @
264e76ca
...
...
@@ -1868,8 +1868,13 @@ class Model(object):
print(params_info)
"""
return
summary
(
self
.
network
,
self
.
_inputs
,
batch_size
,
dtype
)
assert
(
input_size
is
not
None
or
self
.
_inputs
is
not
None
),
"'input_size' or 'self._input' must be set"
if
input_size
is
not
None
:
_input_size
=
input_size
else
:
_input_size
=
self
.
_inputs
return
summary
(
self
.
network
,
_input_size
,
batch_size
,
dtype
)
def
_verify_spec
(
self
,
specs
,
is_input
=
False
):
out_specs
=
[]
...
...
python/paddle/hapi/model_summary.py
浏览文件 @
264e76ca
...
...
@@ -12,7 +12,9 @@
# See the License for the specific language governing permissions and
# limitations under the License.
import
warnings
import
numpy
as
np
import
numbers
import
paddle
import
paddle.nn
as
nn
...
...
@@ -86,8 +88,10 @@ def summary(net, input_size, batch_size=None, dtypes=None):
elif
isinstance
(
input_size
,
list
):
_input_size
=
[]
for
item
in
input_size
:
if
isinstance
(
item
,
int
):
item
=
(
item
,
)
assert
isinstance
(
item
,
(
list
,
InputSpec
)),
'When input_size is list,
\
(
tuple
,
InputSpec
)),
'When input_size is list,
\
expect item in input_size is a tuple or InputSpec, but got {}'
.
format
(
type
(
item
))
...
...
@@ -97,12 +101,19 @@ def summary(net, input_size, batch_size=None, dtypes=None):
batch_size
=
item
.
shape
[
0
]
else
:
_input_size
.
append
(
item
)
elif
isinstance
(
input_size
,
int
):
_input_size
=
(
input_size
,
)
else
:
_input_size
=
input_size
if
batch_size
is
None
:
batch_size
=
-
1
if
not
paddle
.
in_dynamic_mode
():
warnings
.
warn
(
"Your model was created in static mode, this may not get correct summary information!"
)
result
,
params_info
=
summary_string
(
net
,
_input_size
,
batch_size
,
dtypes
)
print
(
result
)
...
...
@@ -117,16 +128,16 @@ def summary_string(model, input_size, batch_size=-1, dtypes=None):
depth
=
len
(
list
(
model
.
sublayers
()))
def
register_hook
(
module
):
def
hook
(
module
,
input
,
output
):
class_name
=
str
(
module
.
__class__
).
split
(
"."
)[
-
1
].
split
(
"'"
)[
0
]
def
register_hook
(
layer
):
def
hook
(
layer
,
input
,
output
):
class_name
=
str
(
layer
.
__class__
).
split
(
"."
)[
-
1
].
split
(
"'"
)[
0
]
try
:
module_idx
=
int
(
module
.
_full_name
.
split
(
'_'
)[
-
1
])
layer_idx
=
int
(
layer
.
_full_name
.
split
(
'_'
)[
-
1
])
except
:
module
_idx
=
len
(
summary
)
layer
_idx
=
len
(
summary
)
m_key
=
"%s-%i"
%
(
class_name
,
module
_idx
+
1
)
m_key
=
"%s-%i"
%
(
class_name
,
layer
_idx
+
1
)
summary
[
m_key
]
=
OrderedDict
()
summary
[
m_key
][
"input_shape"
]
=
list
(
input
[
0
].
shape
)
summary
[
m_key
][
"input_shape"
][
0
]
=
batch_size
...
...
@@ -138,23 +149,50 @@ def summary_string(model, input_size, batch_size=-1, dtypes=None):
summary
[
m_key
][
"output_shape"
][
0
]
=
batch_size
params
=
0
if
hasattr
(
module
,
"weight"
):
params
+=
np
.
prod
(
module
.
weight
.
shape
)
summary
[
m_key
][
"trainable"
]
=
module
.
weight
.
trainable
or
(
not
module
.
weight
.
stop_gradient
)
if
hasattr
(
module
,
"bias"
):
params
+=
np
.
prod
(
module
.
bias
.
shape
)
if
paddle
.
in_dynamic_mode
():
layer_state_dict
=
layer
.
_parameters
else
:
layer_state_dict
=
layer
.
state_dict
()
for
k
,
v
in
layer_state_dict
.
items
():
params
+=
np
.
prod
(
v
.
shape
)
try
:
if
(
getattr
(
getattr
(
layer
,
k
),
'trainable'
))
and
(
not
getattr
(
getattr
(
layer
,
k
),
'stop_gradient'
)):
summary
[
m_key
][
"trainable"
]
=
True
else
:
summary
[
m_key
][
"trainable"
]
=
False
except
:
summary
[
m_key
][
"trainable"
]
=
True
summary
[
m_key
][
"nb_params"
]
=
params
if
(
not
isinstance
(
module
,
nn
.
Sequential
)
and
not
isinstance
(
module
,
nn
.
LayerList
)
and
(
not
(
module
==
model
)
or
depth
<
1
)):
if
(
not
isinstance
(
layer
,
nn
.
Sequential
)
and
not
isinstance
(
layer
,
nn
.
LayerList
)
and
(
not
(
layer
==
model
)
or
depth
<
1
)):
hooks
.
append
(
layer
.
register_forward_post_hook
(
hook
))
def
_check_input_size
(
input_sizes
):
for
input_size
in
input_sizes
:
for
item
in
input_size
:
if
not
isinstance
(
item
,
numbers
.
Number
):
raise
TypeError
(
"Expected item in input size be a number, but got {}"
.
format
(
type
(
item
)))
hooks
.
append
(
module
.
register_forward_post_hook
(
hook
))
if
item
<=
0
:
raise
ValueError
(
"Expected item in input size greater than zero, but got {}"
.
format
(
item
))
if
isinstance
(
input_size
,
tuple
):
input_size
=
[
input_size
]
_check_input_size
(
input_size
)
x
=
[
paddle
.
rand
(
[
2
]
+
list
(
in_size
),
dtype
=
dtype
)
...
...
@@ -193,7 +231,12 @@ def summary_string(model, input_size, batch_size=-1, dtypes=None):
"{0:,}"
.
format
(
summary
[
layer
][
"nb_params"
]),
)
total_params
+=
summary
[
layer
][
"nb_params"
]
total_output
+=
np
.
prod
(
summary
[
layer
][
"output_shape"
])
try
:
total_output
+=
np
.
prod
(
summary
[
layer
][
"output_shape"
])
except
:
for
output_shape
in
summary
[
layer
][
"output_shape"
]:
total_output
+=
np
.
prod
(
output_shape
)
if
"trainable"
in
summary
[
layer
]:
if
summary
[
layer
][
"trainable"
]
==
True
:
trainable_params
+=
summary
[
layer
][
"nb_params"
]
...
...
python/paddle/nn/functional/loss.py
浏览文件 @
264e76ca
...
...
@@ -780,10 +780,10 @@ def kl_div(input, label, reduction='mean', name=None):
input = np.random.uniform(-10, 10, shape).astype('float32')
target = np.random.uniform(-10, 10, shape).astype('float32')
# 'batchmean' reduction, loss shape will be [
N
]
# 'batchmean' reduction, loss shape will be [
1
]
pred_loss = F.kl_div(paddle.to_tensor(input),
paddle.to_tensor(target), reduction='batchmean')
# shape=[
5
]
# shape=[
1
]
# 'mean' reduction, loss shape will be [1]
pred_loss = F.kl_div(paddle.to_tensor(input),
...
...
python/paddle/nn/layer/conv.py
浏览文件 @
264e76ca
...
...
@@ -1084,7 +1084,7 @@ class ConvTranspose3d(_ConvNd):
bias_attr
=
bias_attr
,
data_format
=
data_format
)
def
forward
(
self
,
x
,
output_size
):
def
forward
(
self
,
x
,
output_size
=
None
):
if
output_size
is
None
:
output_padding
=
self
.
output_padding
else
:
...
...
python/paddle/nn/layer/loss.py
浏览文件 @
264e76ca
...
...
@@ -627,10 +627,13 @@ class KLDivLoss(fluid.dygraph.Layer):
$$l(x, y) = y * (\log(y) - x)$$
Parameters:
reduction (str, optional): Indicate how to average the loss,
the candicates are ``'none'`` | ``'mean'`` | ``'sum'``.
If :attr:`reduction` is ``'mean'``, the reduced mean loss is returned;
Default is ``'mean'``.
reduction (Tensor): Indicate how to average the loss,
the candicates are ``'none'`` | ``'batchmean'`` | ``'mean'`` | ``'sum'``.
If `reduction` is ``'mean'``, the reduced mean loss is returned;
If `reduction` is ``'batchmean'``, the sum loss divided by batch size is returned;
if `reduction` is ``'sum'``, the reduced sum loss is returned;
if `reduction` is ``'none'``, no reduction will be apllied.
Default is ``'mean'``.
Shape:
...
...
@@ -654,11 +657,11 @@ class KLDivLoss(fluid.dygraph.Layer):
x = np.random.uniform(-10, 10, shape).astype('float32')
target = np.random.uniform(-10, 10, shape).astype('float32')
# 'batchmean' reduction, loss shape will be [
N
]
# 'batchmean' reduction, loss shape will be [
1
]
kldiv_criterion = nn.KLDivLoss(reduction='batchmean')
pred_loss = kldiv_criterion(paddle.to_tensor(x),
paddle.to_tensor(target))
# shape=[
5
]
# shape=[
1
]
# 'mean' reduction, loss shape will be [1]
kldiv_criterion = nn.KLDivLoss(reduction='mean')
...
...
@@ -684,7 +687,7 @@ class KLDivLoss(fluid.dygraph.Layer):
self
.
reduction
=
reduction
def
forward
(
self
,
input
,
label
):
out
=
paddle
.
nn
.
functional
.
kl_div
(
input
,
label
,
self
.
reduction
)
out
=
F
.
kl_div
(
input
,
label
,
self
.
reduction
)
return
out
...
...
python/paddle/tests/test_model.py
浏览文件 @
264e76ca
...
...
@@ -519,6 +519,28 @@ class TestModelFunction(unittest.TestCase):
np
.
testing
.
assert_allclose
(
params_info
[
'total_params'
],
gt_params
)
print
(
params_info
)
model
.
summary
(
input_size
=
(
20
))
model
.
summary
(
input_size
=
[(
20
)])
model
.
summary
(
input_size
=
(
20
),
batch_size
=
2
)
def
test_summary_nlp
(
self
):
paddle
.
enable_static
()
nlp_net
=
paddle
.
nn
.
GRU
(
input_size
=
2
,
hidden_size
=
3
,
num_layers
=
3
)
paddle
.
summary
(
nlp_net
,
(
1
,
2
))
def
test_summary_error
(
self
):
with
self
.
assertRaises
(
TypeError
):
nlp_net
=
paddle
.
nn
.
GRU
(
input_size
=
2
,
hidden_size
=
3
,
num_layers
=
3
)
paddle
.
summary
(
nlp_net
,
(
1
,
'2'
))
with
self
.
assertRaises
(
ValueError
):
nlp_net
=
paddle
.
nn
.
GRU
(
input_size
=
2
,
hidden_size
=
3
,
num_layers
=
3
)
paddle
.
summary
(
nlp_net
,
(
-
1
,
-
1
))
paddle
.
disable_static
()
nlp_net
=
paddle
.
nn
.
GRU
(
input_size
=
2
,
hidden_size
=
3
,
num_layers
=
3
)
paddle
.
summary
(
nlp_net
,
(
1
,
2
))
def
test_export_deploy_model
(
self
):
for
dynamic
in
[
True
,
False
]:
fluid
.
enable_dygraph
()
if
dynamic
else
None
...
...
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