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
Serving
提交
4cab45e7
S
Serving
项目概览
PaddlePaddle
/
Serving
12 个月 前同步成功
通知
183
Star
833
Fork
253
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
105
列表
看板
标记
里程碑
合并请求
10
Wiki
2
Wiki
分析
仓库
DevOps
项目成员
Pages
S
Serving
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
105
Issue
105
列表
看板
标记
里程碑
合并请求
10
合并请求
10
Pages
分析
分析
仓库分析
DevOps
Wiki
2
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 >>
提交
4cab45e7
编写于
7月 30, 2019
作者:
W
wangguibao
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
CTR prediction serving
上级
416ecd8f
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
33 addition
and
23 deletion
+33
-23
demo-serving/op/ctr_prediction_op.cpp
demo-serving/op/ctr_prediction_op.cpp
+32
-22
demo-serving/proto/ctr_prediction.proto
demo-serving/proto/ctr_prediction.proto
+1
-1
未找到文件。
demo-serving/op/ctr_prediction_op.cpp
浏览文件 @
4cab45e7
...
...
@@ -28,6 +28,8 @@ using baidu::paddle_serving::predictor::ctr_prediction::Response;
using
baidu
::
paddle_serving
::
predictor
::
ctr_prediction
::
CTRReqInstance
;
using
baidu
::
paddle_serving
::
predictor
::
ctr_prediction
::
Request
;
const
int
VARIABLE_NAME_LEN
=
256
;
// Total 26 sparse input + 1 dense input
const
int
CTR_PREDICTION_INPUT_SLOTS
=
27
;
...
...
@@ -44,8 +46,8 @@ struct CubeValue {
int
error
;
std
::
string
buff
;
};
#endif
void
fill_response_with_message
(
Response
*
response
,
int
err_code
,
std
::
string
err_msg
)
{
...
...
@@ -69,11 +71,11 @@ int CTRPredictionOp::inference() {
if
(
sample_size
<=
0
)
{
LOG
(
WARNING
)
<<
"No instances need to inference!"
;
fill_response_with_message
(
res
,
-
1
,
"Sample size invalid"
);
return
-
1
;
return
0
;
}
paddle
::
PaddleTensor
lod_tensors
[
CTR_PREDICTION_INPUT_SLOTS
];
for
(
int
i
=
0
;
i
<
CTR_PREDICTION_
SPARSE
_SLOTS
;
++
i
)
{
for
(
int
i
=
0
;
i
<
CTR_PREDICTION_
INPUT
_SLOTS
;
++
i
)
{
lod_tensors
[
i
].
dtype
=
paddle
::
PaddleDType
::
FLOAT32
;
std
::
vector
<
std
::
vector
<
size_t
>>
&
lod
=
lod_tensors
[
i
].
lod
;
lod
.
resize
(
1
);
...
...
@@ -86,11 +88,11 @@ int CTRPredictionOp::inference() {
for
(
uint32_t
si
=
0
;
si
<
sample_size
;
++
si
)
{
const
CTRReqInstance
&
req_instance
=
req
->
instances
(
si
);
if
(
req_instance
.
sparse_ids_size
()
!=
CTR_PREDICTION_
DENSE_DIM
)
{
if
(
req_instance
.
sparse_ids_size
()
!=
CTR_PREDICTION_
SPARSE_SLOTS
)
{
std
::
ostringstream
iss
;
iss
<<
"
dense input size != "
<<
CTR_PREDICTION_DENSE_DIM
;
iss
<<
"
Sparse input size != "
<<
CTR_PREDICTION_SPARSE_SLOTS
;
fill_response_with_message
(
res
,
-
1
,
iss
.
str
());
return
-
1
;
return
0
;
}
for
(
int
i
=
0
;
i
<
req_instance
.
sparse_ids_size
();
++
i
)
{
...
...
@@ -110,16 +112,22 @@ int CTRPredictionOp::inference() {
float
buff
[
CTR_PREDICTION_EMBEDDING_SIZE
]
=
{
0.01
,
0.02
,
0.03
,
0.04
,
0.05
,
0.06
,
0.07
,
0.08
,
0.09
,
0.00
};
for
(
int
i
=
0
;
i
<
keys
.
size
();
++
i
)
{
values
[
i
].
error
=
0
;
values
[
i
].
buff
=
std
::
string
(
reinterpret_cast
<
char
*>
(
buff
),
sizeof
(
buff
));
CubeValue
value
;
value
.
error
=
0
;
value
.
buff
=
std
::
string
(
reinterpret_cast
<
char
*>
(
buff
),
sizeof
(
buff
));
values
.
push_back
(
value
);
}
#endif
// Sparse embeddings
for
(
int
i
=
0
;
i
<
CTR_PREDICTION_SPARSE_SLOTS
;
++
i
)
{
paddle
::
PaddleTensor
lod_tensor
=
lod_tensors
[
i
];
paddle
::
PaddleTensor
&
lod_tensor
=
lod_tensors
[
i
];
std
::
vector
<
std
::
vector
<
size_t
>>
&
lod
=
lod_tensor
.
lod
;
char
name
[
VARIABLE_NAME_LEN
];
snprintf
(
name
,
VARIABLE_NAME_LEN
,
"embedding_%d.tmp_0"
,
i
);
lod_tensor
.
name
=
std
::
string
(
name
);
for
(
uint32_t
si
=
0
;
si
<
sample_size
;
++
si
)
{
const
CTRReqInstance
&
req_instance
=
req
->
instances
(
si
);
lod
[
0
].
push_back
(
lod
[
0
].
back
()
+
1
);
...
...
@@ -140,7 +148,7 @@ int CTRPredictionOp::inference() {
LOG
(
ERROR
)
<<
"Embedding vector size not expected"
;
fill_response_with_message
(
res
,
-
1
,
"Embedding vector size not expected"
);
return
-
1
;
return
0
;
}
memcpy
(
data_ptr
,
values
[
idx
].
buff
.
data
(),
values
[
idx
].
buff
.
size
());
...
...
@@ -151,9 +159,10 @@ int CTRPredictionOp::inference() {
}
// Dense features
paddle
::
PaddleTensor
lod_tensor
=
lod_tensors
[
CTR_PREDICTION_DENSE_SLOT_ID
];
lod_tensor
.
dtype
=
paddle
::
PaddleDType
::
INT64
;
paddle
::
PaddleTensor
&
lod_tensor
=
lod_tensors
[
CTR_PREDICTION_DENSE_SLOT_ID
];
lod_tensor
.
dtype
=
paddle
::
PaddleDType
::
FLOAT32
;
std
::
vector
<
std
::
vector
<
size_t
>>
&
lod
=
lod_tensor
.
lod
;
lod_tensor
.
name
=
std
::
string
(
"dense_input"
);
for
(
uint32_t
si
=
0
;
si
<
sample_size
;
++
si
)
{
const
CTRReqInstance
&
req_instance
=
req
->
instances
(
si
);
...
...
@@ -161,22 +170,23 @@ int CTRPredictionOp::inference() {
std
::
ostringstream
iss
;
iss
<<
"dense input size != "
<<
CTR_PREDICTION_DENSE_DIM
;
fill_response_with_message
(
res
,
-
1
,
iss
.
str
());
return
-
1
;
return
0
;
}
lod
[
0
].
push_back
(
lod
[
0
].
back
()
+
req_instance
.
dense_ids_size
());
}
lod_tensor
.
shape
=
{
lod
[
0
].
back
(),
CTR_PREDICTION_DENSE_DIM
};
lod_tensor
.
data
.
Resize
(
lod
[
0
].
back
()
*
sizeof
(
int64_t
));
lod_tensor
.
shape
=
{
lod
[
0
].
back
()
/
CTR_PREDICTION_DENSE_DIM
,
CTR_PREDICTION_DENSE_DIM
};
lod_tensor
.
data
.
Resize
(
lod
[
0
].
back
()
*
sizeof
(
float
));
int
offset
=
0
;
for
(
uint32_t
si
=
0
;
si
<
sample_size
;
++
si
)
{
int64_t
*
data_ptr
=
static_cast
<
int64_
t
*>
(
lod_tensor
.
data
.
data
())
+
offset
;
float
*
data_ptr
=
static_cast
<
floa
t
*>
(
lod_tensor
.
data
.
data
())
+
offset
;
const
CTRReqInstance
&
req_instance
=
req
->
instances
(
si
);
int
id_count
=
req_instance
.
dense_ids_size
();
memcpy
(
data_ptr
,
req_instance
.
dense_ids
().
data
(),
sizeof
(
int64_
t
)
*
req_instance
.
dense_ids_size
());
sizeof
(
floa
t
)
*
req_instance
.
dense_ids_size
());
offset
+=
req_instance
.
dense_ids_size
();
}
...
...
@@ -186,7 +196,7 @@ int CTRPredictionOp::inference() {
if
(
!
out
)
{
LOG
(
ERROR
)
<<
"Failed get tls output object"
;
fill_response_with_message
(
res
,
-
1
,
"Failed get thread local resource"
);
return
-
1
;
return
0
;
}
// call paddle fluid model for inferencing
...
...
@@ -195,13 +205,13 @@ int CTRPredictionOp::inference() {
LOG
(
ERROR
)
<<
"Failed do infer in fluid model: "
<<
CTR_PREDICTION_MODEL_NAME
;
fill_response_with_message
(
res
,
-
1
,
"Failed do infer in fluid model"
);
return
-
1
;
return
0
;
}
if
(
out
->
size
()
!=
in
->
size
()
)
{
if
(
out
->
size
()
!=
sample_size
)
{
LOG
(
ERROR
)
<<
"Output tensor size not equal that of input"
;
fill_response_with_message
(
res
,
-
1
,
"Output size != input size"
);
return
-
1
;
return
0
;
}
for
(
size_t
i
=
0
;
i
<
out
->
size
();
++
i
)
{
...
...
@@ -211,7 +221,7 @@ int CTRPredictionOp::inference() {
if
(
out
->
at
(
i
).
dtype
!=
paddle
::
PaddleDType
::
FLOAT32
)
{
LOG
(
ERROR
)
<<
"Expected data type float"
;
fill_response_with_message
(
res
,
-
1
,
"Expected data type float"
);
return
-
1
;
return
0
;
}
float
*
data
=
static_cast
<
float
*>
(
out
->
at
(
i
).
data
.
data
());
...
...
demo-serving/proto/ctr_prediction.proto
浏览文件 @
4cab45e7
...
...
@@ -21,7 +21,7 @@ option cc_generic_services = true;
message
CTRReqInstance
{
repeated
int64
sparse_ids
=
1
;
repeated
int64
dense_ids
=
2
;
repeated
float
dense_ids
=
2
;
};
message
Request
{
repeated
CTRReqInstance
instances
=
1
;
};
...
...
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