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
c6f173b0
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
c6f173b0
编写于
9月 13, 2022
作者:
C
Charles-hit
提交者:
GitHub
9月 13, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
support tile op backward refuse forward (#45942)
上级
1eefd66a
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
79 addition
and
5 deletion
+79
-5
paddle/phi/api/yaml/legacy_backward.yaml
paddle/phi/api/yaml/legacy_backward.yaml
+1
-4
python/paddle/fluid/tests/unittests/test_tile_op.py
python/paddle/fluid/tests/unittests/test_tile_op.py
+78
-1
未找到文件。
paddle/phi/api/yaml/legacy_backward.yaml
浏览文件 @
c6f173b0
...
...
@@ -2530,10 +2530,7 @@
forward
:
tile_grad (Tensor x, Tensor grad_out, IntArray repeat_times) -> Tensor(grad_x)
args
:
(Tensor grad_x_grad, IntArray repeat_times)
output
:
Tensor(grad_out_grad)
infer_meta
:
func
:
TileInferMeta
kernel
:
func
:
tile
invoke
:
tile(grad_x_grad, repeat_times)
-
backward_api
:
tile_grad
forward
:
tile (Tensor x, IntArray repeat_times) -> Tensor(out)
...
...
python/paddle/fluid/tests/unittests/test_tile_op.py
浏览文件 @
c6f173b0
...
...
@@ -19,7 +19,10 @@ import numpy as np
from
op_test
import
OpTest
import
paddle
import
paddle.fluid
as
fluid
from
paddle.fluid
import
compiler
,
Program
,
program_guard
from
paddle.fluid
import
compiler
,
Program
,
program_guard
,
core
import
gradient_checker
from
decorator_helper
import
prog_scope
import
paddle.fluid.layers
as
layers
#Situation 1: repeat_times is a list (without tensor)
...
...
@@ -263,6 +266,80 @@ class TestTileAPI(unittest.TestCase):
assert
np
.
array_equal
(
out_3
.
numpy
(),
np
.
tile
(
np_x
,
(
2
,
3
)))
class
TestTileDoubleGradCheck
(
unittest
.
TestCase
):
def
tile_wrapper
(
self
,
x
):
return
paddle
.
tile
(
x
[
0
],
[
2
,
1
])
@
prog_scope
()
def
func
(
self
,
place
):
# the shape of input variable should be clearly specified, not inlcude -1.
eps
=
0.005
dtype
=
np
.
float32
data
=
layers
.
data
(
'data'
,
[
1
,
2
],
False
,
dtype
)
data
.
persistable
=
True
out
=
paddle
.
tile
(
data
,
[
2
,
1
])
data_arr
=
np
.
random
.
uniform
(
-
1
,
1
,
data
.
shape
).
astype
(
dtype
)
gradient_checker
.
double_grad_check
([
data
],
out
,
x_init
=
[
data_arr
],
place
=
place
,
eps
=
eps
)
fluid
.
set_flags
({
"FLAGS_retain_grad_for_all_tensor"
:
True
})
gradient_checker
.
double_grad_check_for_dygraph
(
self
.
tile_wrapper
,
[
data
],
out
,
x_init
=
[
data_arr
],
place
=
place
)
def
test_grad
(
self
):
paddle
.
enable_static
()
places
=
[
fluid
.
CPUPlace
()]
if
core
.
is_compiled_with_cuda
():
places
.
append
(
fluid
.
CUDAPlace
(
0
))
for
p
in
places
:
self
.
func
(
p
)
class
TestTileTripleGradCheck
(
unittest
.
TestCase
):
def
tile_wrapper
(
self
,
x
):
return
paddle
.
tile
(
x
[
0
],
[
2
,
1
])
@
prog_scope
()
def
func
(
self
,
place
):
# the shape of input variable should be clearly specified, not inlcude -1.
eps
=
0.005
dtype
=
np
.
float32
data
=
layers
.
data
(
'data'
,
[
1
,
2
],
False
,
dtype
)
data
.
persistable
=
True
out
=
paddle
.
tile
(
data
,
[
2
,
1
])
data_arr
=
np
.
random
.
uniform
(
-
1
,
1
,
data
.
shape
).
astype
(
dtype
)
gradient_checker
.
triple_grad_check
([
data
],
out
,
x_init
=
[
data_arr
],
place
=
place
,
eps
=
eps
)
fluid
.
set_flags
({
"FLAGS_retain_grad_for_all_tensor"
:
True
})
gradient_checker
.
triple_grad_check_for_dygraph
(
self
.
tile_wrapper
,
[
data
],
out
,
x_init
=
[
data_arr
],
place
=
place
)
def
test_grad
(
self
):
paddle
.
enable_static
()
places
=
[
fluid
.
CPUPlace
()]
if
core
.
is_compiled_with_cuda
():
places
.
append
(
fluid
.
CUDAPlace
(
0
))
for
p
in
places
:
self
.
func
(
p
)
if
__name__
==
"__main__"
:
paddle
.
enable_static
()
unittest
.
main
()
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