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
4e3f0b95
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 >>
未验证
提交
4e3f0b95
编写于
8月 24, 2022
作者:
W
Wang Bojun
提交者:
GitHub
8月 24, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
fix mean/variance shape infer bug during loop call of dynamic trt enqueue (#45387)
* fix bug fix
上级
73e41c89
变更
1
显示空白变更内容
内联
并排
Showing
1 changed file
with
16 addition
and
5 deletion
+16
-5
paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.cu
...e/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.cu
+16
-5
未找到文件。
paddle/fluid/inference/tensorrt/plugin/layer_norm_op_plugin.cu
浏览文件 @
4e3f0b95
...
...
@@ -177,9 +177,20 @@ int LayerNormPluginDynamic::enqueue(
}
// in dynamic shape
// the batch num should be involved in mean/variance shape
mean_shape_
[
0
]
*=
input_dims
.
d
[
0
];
variance_shape_
[
0
]
*=
input_dims
.
d
[
0
];
PADDLE_ENFORCE_EQ
(
1
,
mean_shape_
.
size
(),
platform
::
errors
::
InvalidArgument
(
"Size of mean_shape vector should be equal to 1,"
"but got Size of mean_shape vector:%d"
,
mean_shape_
.
size
()));
PADDLE_ENFORCE_EQ
(
1
,
variance_shape_
.
size
(),
platform
::
errors
::
InvalidArgument
(
"Size of variance_shape vector should be equal to 1,"
"but got Size of mean_shape vector:%d"
,
mean_shape_
.
size
()));
int64_t
batched_mean_shape
=
mean_shape_
[
0
]
*
input_dims
.
d
[
0
];
int64_t
batched_variance_shape
=
variance_shape_
[
0
]
*
input_dims
.
d
[
0
];
const
auto
input_ddim
=
phi
::
make_ddim
(
input_shape
);
auto
matrix_dim
=
phi
::
flatten_to_2d
(
input_ddim
,
begin_norm_axis
);
int
feature_size
=
static_cast
<
int
>
(
matrix_dim
[
1
]);
...
...
@@ -206,8 +217,8 @@ int LayerNormPluginDynamic::enqueue(
float
*
output
=
static_cast
<
float
*>
(
outputs
[
0
]);
scale_t
.
Resize
(
phi
::
make_ddim
({
feature_size
}));
bias_t
.
Resize
(
phi
::
make_ddim
({
feature_size
}));
mean_t
.
Resize
(
phi
::
make_ddim
(
mean_shape_
));
variance_t
.
Resize
(
phi
::
make_ddim
(
variance_shape_
));
mean_t
.
Resize
(
phi
::
make_ddim
(
{
batched_mean_shape
}
));
variance_t
.
Resize
(
phi
::
make_ddim
(
{
batched_variance_shape
}
));
float
*
scale_d
=
scale_t
.
mutable_data
<
float
>
(
platform
::
CUDAPlace
(
device_id
));
...
...
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