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
wk1234123
Paddle
提交
9ed16a43
P
Paddle
项目概览
wk1234123
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
4
Star
0
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 >>
未验证
提交
9ed16a43
编写于
6月 15, 2020
作者:
Y
Yiqun Liu
提交者:
GitHub
6月 15, 2020
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Fix random fail because of precision problem in unittest of fusion_group (#25051)
上级
bef4afa6
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
20 addition
and
13 deletion
+20
-13
python/paddle/fluid/tests/unittests/ir/pass_test.py
python/paddle/fluid/tests/unittests/ir/pass_test.py
+13
-5
python/paddle/fluid/tests/unittests/ir/test_ir_fusion_group_pass.py
...dle/fluid/tests/unittests/ir/test_ir_fusion_group_pass.py
+7
-8
未找到文件。
python/paddle/fluid/tests/unittests/ir/pass_test.py
浏览文件 @
9ed16a43
...
...
@@ -148,11 +148,19 @@ class PassTest(unittest.TestCase):
"Checking the number of fetchs failed. Expected: {}, Received: {}"
.
format
(
len
(
self
.
fetch_list
),
len
(
outs_opt
)))
for
i
in
six
.
moves
.
xrange
(
len
(
self
.
fetch_list
)):
self
.
assertTrue
(
np
.
allclose
(
outs_opt
[
i
],
outs
[
i
],
atol
=
atol
),
"Output < {} > has diff at {}, expected {} but got {}"
.
format
(
self
.
fetch_list
[
i
],
str
(
place
),
outs_opt
[
i
],
outs
[
i
]))
is_allclose
=
np
.
allclose
(
outs_opt
[
i
],
outs
[
i
],
atol
=
atol
)
if
not
is_allclose
:
a
=
outs_opt
[
i
]
b
=
outs
[
i
]
diff_mat
=
np
.
abs
(
a
-
b
)
/
np
.
abs
(
a
)
max_diff
=
np
.
max
(
diff_mat
)
offset
=
np
.
argmax
(
diff_mat
>
atol
)
self
.
assertTrue
(
is_allclose
,
"Output (name: %s, shape: %s, dtype: %s) has diff at %s. The maximum diff is %e, first error element is %d, expected %e, but got %e"
%
(
self
.
fetch_list
[
i
].
name
,
str
(
self
.
fetch_list
[
i
].
shape
),
self
.
fetch_list
[
i
].
dtype
,
str
(
place
),
max_diff
,
offset
,
a
.
flatten
()[
offset
],
b
.
flatten
()[
offset
]))
def
_check_fused_ops
(
self
,
program
):
'''
...
...
python/paddle/fluid/tests/unittests/ir/test_ir_fusion_group_pass.py
浏览文件 @
9ed16a43
...
...
@@ -132,12 +132,17 @@ class FusionGroupPassTestCastAndFP16(FusionGroupPassTest):
# subgraph with 2 op nodes
tmp_0
=
self
.
feed_vars
[
0
]
*
self
.
feed_vars
[
1
]
tmp_1
=
layers
.
softmax
(
layers
.
cast
(
tmp_0
,
dtype
=
"float16"
))
tmp_2
=
layers
.
mul
(
tmp_0
,
self
.
feed_vars
[
2
])
tmp_1
=
layers
.
cast
(
tmp_0
,
dtype
=
"float16"
)
zero
=
layers
.
fill_constant
(
shape
=
[
128
],
dtype
=
"float16"
,
value
=
0
)
# TODO(xreki): fix precision problem when using softmax of float16.
# tmp_2 = layers.softmax(tmp_1)
tmp_2
=
layers
.
elementwise_add
(
tmp_1
,
zero
)
tmp_3
=
layers
.
mul
(
tmp_0
,
self
.
feed_vars
[
2
])
# subgraph with 4 op nodes
tmp_3
=
layers
.
cast
(
tmp_2
,
dtype
=
"float16"
)
tmp_4
=
layers
.
relu
(
tmp_1
+
tmp_3
)
tmp_5
=
layers
.
cast
(
tmp_4
,
dtype
=
dtype
)
tmp_3
=
layers
.
cast
(
tmp_2
,
dtype
=
dtype
)
self
.
append_gradients
(
tmp_5
)
...
...
@@ -204,12 +209,6 @@ class FusionGroupPassFillConstantTest(FusionGroupPassTest):
self
.
num_fused_ops
=
1
self
.
fetch_list
=
[
tmp_2
,
self
.
grad
(
tmp_0
)]
def
setUp
(
self
):
self
.
build_program
(
"float32"
)
self
.
feeds
=
self
.
_feed_random_data
(
self
.
feed_vars
)
self
.
pass_names
=
"fusion_group_pass"
self
.
fused_op_type
=
"fusion_group"
if
__name__
==
"__main__"
:
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