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
BaiXuePrincess
Paddle
提交
c384b00d
P
Paddle
项目概览
BaiXuePrincess
/
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
未验证
提交
c384b00d
编写于
9月 20, 2022
作者:
Z
zhoutianzi666
提交者:
GitHub
9月 20, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
[Inference] fix preln_residual_bias_fuse_pass bug in TNT_small model (#46178) (#46260)
* fix preln_residual_bias_fuse_pass bug in TNT_small model
上级
4f28a4c2
变更
1
显示空白变更内容
内联
并排
Showing
1 changed file
with
16 addition
and
4 deletion
+16
-4
paddle/fluid/framework/ir/preln_residual_bias_fuse_pass.cc
paddle/fluid/framework/ir/preln_residual_bias_fuse_pass.cc
+16
-4
未找到文件。
paddle/fluid/framework/ir/preln_residual_bias_fuse_pass.cc
浏览文件 @
c384b00d
...
@@ -143,10 +143,7 @@ void PrelnResidualBiasFusePass::ApplyImpl(ir::Graph *graph) const {
...
@@ -143,10 +143,7 @@ void PrelnResidualBiasFusePass::ApplyImpl(ir::Graph *graph) const {
LOG
(
WARNING
)
<<
"The subgraph is empty."
;
LOG
(
WARNING
)
<<
"The subgraph is empty."
;
return
;
return
;
}
}
if
(
!
IsCompat
(
subgraph
,
graph
))
{
LOG
(
WARNING
)
<<
"preln_residual_bias pass in op compat failed."
;
return
;
}
VLOG
(
4
)
<<
"handle PrelnResidualBias fuse"
;
VLOG
(
4
)
<<
"handle PrelnResidualBias fuse"
;
GET_IR_NODE_FROM_SUBGRAPH
(
GET_IR_NODE_FROM_SUBGRAPH
(
elementwise_bias
,
elementwise_bias
,
fused_pattern
);
elementwise_bias
,
elementwise_bias
,
fused_pattern
);
...
@@ -164,6 +161,21 @@ void PrelnResidualBiasFusePass::ApplyImpl(ir::Graph *graph) const {
...
@@ -164,6 +161,21 @@ void PrelnResidualBiasFusePass::ApplyImpl(ir::Graph *graph) const {
GET_IR_NODE_FROM_SUBGRAPH
(
layer_norm_mean
,
layer_norm_mean
,
fused_pattern
);
GET_IR_NODE_FROM_SUBGRAPH
(
layer_norm_mean
,
layer_norm_mean
,
fused_pattern
);
GET_IR_NODE_FROM_SUBGRAPH
(
GET_IR_NODE_FROM_SUBGRAPH
(
layer_norm_variance
,
layer_norm_variance
,
fused_pattern
);
layer_norm_variance
,
layer_norm_variance
,
fused_pattern
);
// We can not accept that two or more layer_norm is connected to
// elementwise1_out. This will lead to two or more PrelnResidualBias
// patterns is found near elementwise1_out, and these patterns will interact
// on each other, so we make below check to ensure only one
// PrelnResidualBias pattern is delalted with.
for
(
auto
op
:
elementwise1_out
->
inputs
)
{
if
(
op
->
Name
()
==
"preln_residual_bias"
)
return
;
}
if
(
!
IsCompat
(
subgraph
,
graph
))
{
LOG
(
WARNING
)
<<
"preln_residual_bias pass in op compat failed."
;
return
;
}
std
::
unordered_set
<
const
Node
*>
del_node_set
;
std
::
unordered_set
<
const
Node
*>
del_node_set
;
// Create an PrelnResidualBias op node
// Create an PrelnResidualBias op node
OpDesc
new_desc
;
OpDesc
new_desc
;
...
...
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