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
7cebec4b
P
Paddle
项目概览
PaddlePaddle
/
Paddle
12 个月 前同步成功
通知
2292
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 >>
提交
7cebec4b
编写于
6月 10, 2018
作者:
Q
qiaolongfei
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
init merge_ids_op
上级
d896134f
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
180 addition
and
0 deletion
+180
-0
paddle/fluid/operators/merge_ids_op.cc
paddle/fluid/operators/merge_ids_op.cc
+97
-0
paddle/fluid/operators/merge_ids_op.h
paddle/fluid/operators/merge_ids_op.h
+83
-0
未找到文件。
paddle/fluid/operators/merge_ids_op.cc
0 → 100644
浏览文件 @
7cebec4b
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/fluid/operators/merge_ids_op.h"
namespace
paddle
{
namespace
operators
{
class
MergeIdsOpMaker
:
public
framework
::
OpProtoAndCheckerMaker
{
public:
void
Make
()
override
{
AddInput
(
"Ids"
,
"(LoDTensor) the input ids with shape{batch_num, 1}"
);
AddInput
(
"X"
,
"(LoDTensor) the input tensor with shape{batch_num, N}, N is the "
"size of embedding table"
)
.
AsDuplicable
();
AddOutput
(
"Out"
,
"(LoDTensor) The merged outputs of the input tensors."
);
AddComment
(
R"DOC(
Merge multi LoDTensor's into one according to Ids's shard num.
The values in the input LoDTensor are lookuped from the output of splite_ids_op
Example:
Input:
Ids = [1,2,3,4,5,6]
X0 = [[0.1 0.2] # 3
[0.2 0.3]] # 6
X1 = [[0.3 0.4] # 1
[0.4 0.5]] # 4
X2 = [[0.5 0.6] # 2
[0.6 0.7]] # 5
Output:
Out = [[0.3 0.4] # 1
[0.5 0.6] # 2
[0.1 0.2] # 3
[0.4 0.5] # 4
[0.6 0.7] # 5
[0.2 0.3]] # 6
)DOC"
);
}
};
class
MergeIdsOp
:
public
framework
::
OperatorWithKernel
{
public:
using
framework
::
OperatorWithKernel
::
OperatorWithKernel
;
void
InferShape
(
framework
::
InferShapeContext
*
ctx
)
const
override
{
PADDLE_ENFORCE
(
ctx
->
HasInput
(
"Ids"
),
"MergeIdsOp must has input Ids."
);
PADDLE_ENFORCE
(
ctx
->
HasInputs
(
"X"
),
"MergeIdsOp must has input X."
);
PADDLE_ENFORCE
(
ctx
->
HasOutput
(
"Out"
),
"MergeIdsOp must has output Out."
);
auto
ids_var_type
=
ctx
->
GetInputsVarType
(
"Ids"
).
front
();
auto
ids_dims
=
ctx
->
GetInputDim
(
"Ids"
);
if
(
ids_var_type
==
framework
::
proto
::
VarType
::
LOD_TENSOR
)
{
PADDLE_ENFORCE_EQ
(
ids_dims
.
size
(),
2
);
PADDLE_ENFORCE_EQ
(
ids_dims
[
1
],
1
);
}
auto
x_var_type
=
ctx
->
GetInputsVarType
(
"X"
);
for
(
auto
&
var_type
:
x_var_type
)
{
PADDLE_ENFORCE_EQ
(
var_type
,
framework
::
proto
::
VarType
::
LOD_TENSOR
,
"input X only support lod tensors"
);
}
ctx
->
ShareLoD
(
"Ids"
,
"Out"
);
}
};
class
MergeIdsOpInferVarType
:
public
framework
::
VarTypeInference
{
public:
void
operator
()(
const
framework
::
OpDesc
&
op_desc
,
framework
::
BlockDesc
*
block
)
const
override
{
auto
*
input_var
=
block
->
Var
(
op_desc
.
Input
(
"Ids"
)[
0
]);
for
(
auto
&
out_var
:
op_desc
.
Output
(
"Out"
))
{
block
->
Var
(
out_var
)
->
SetType
(
input_var
->
GetType
());
}
}
};
}
// namespace operators
}
// namespace paddle
namespace
ops
=
paddle
::
operators
;
REGISTER_OPERATOR
(
merge_ids
,
ops
::
MergeIdsOp
,
ops
::
MergeIdsOpMaker
,
ops
::
MergeIdsOpInferVarType
);
REGISTER_OP_CPU_KERNEL
(
merge_ids
,
ops
::
MergeIdsOpKernel
<
paddle
::
platform
::
CPUPlace
,
int64_t
>
,
ops
::
MergeIdsOpKernel
<
paddle
::
platform
::
CPUPlace
,
float
>
);
paddle/fluid/operators/merge_ids_op.h
0 → 100644
浏览文件 @
7cebec4b
/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include <vector>
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/tensor_util.h"
#include "paddle/fluid/operators/math/selected_rows_functor.h"
namespace
paddle
{
namespace
operators
{
template
<
typename
DeviceContext
,
typename
T
>
class
MergeIdsOpKernel
:
public
framework
::
OpKernel
<
T
>
{
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
auto
place
=
ctx
.
GetPlace
();
if
(
!
platform
::
is_cpu_place
(
place
))
{
PADDLE_THROW
(
"MergeIds do not support GPU kernel"
);
}
const
auto
*
ids_var
=
ctx
.
InputVar
(
"Ids"
);
PADDLE_ENFORCE
(
ids_var
->
IsType
<
framework
::
LoDTensor
>
(),
"only support to merge Ids of LoDTensor"
);
const
auto
&
ids_tensor
=
ids_var
->
Get
<
framework
::
LoDTensor
>
();
const
auto
&
ids_dims
=
ids_tensor
.
dims
();
const
T
*
ids
=
ids_tensor
.
data
<
T
>
();
auto
x_tensors
=
ctx
.
MultiInput
<
framework
::
LoDTensor
>
(
"X"
);
auto
*
out
=
ctx
.
Output
<
framework
::
LoDTensor
>
(
"Out"
);
int
batch_size
=
0
;
int
embedding_size
=
0
;
for
(
auto
&
input
:
x_tensors
)
{
if
(
embedding_size
==
0
)
{
embedding_size
=
input
->
dims
()[
1
];
}
PADDLE_ENFORCE_EQ
(
embedding_size
,
input
->
dims
()[
1
],
"embedding size of all input should be the same"
);
batch_size
+=
input
->
dims
()[
0
];
}
PADDLE_ENFORCE_EQ
(
batch_size
,
ids_dims
[
0
],
"the batch size of ids and embedding value should be the same"
);
const
size_t
shard_num
=
x_tensors
.
size
();
if
(
shard_num
==
1
)
{
VLOG
(
3
)
<<
"only one shard, we can copy the data directly"
;
TensorCopy
(
ids_tensor
,
place
,
out
);
}
else
{
std
::
vector
<
int
>
in_indexs
(
shard_num
,
0
);
auto
*
out_data
=
out
->
mutable_data
<
T
>
(
ids_dims
,
place
);
// copy data from ins[shard_num] to out.
for
(
int
i
=
0
;
i
<
ids_dims
[
0
];
++
i
)
{
T
id
=
ids
[
i
];
size_t
shard_id
=
static_cast
<
size_t
>
(
id
)
%
shard_num
;
int
index
=
in_indexs
[
shard_id
];
memcpy
(
out_data
+
embedding_size
*
i
,
x_tensors
[
shard_id
]
->
data
<
T
>
()
+
index
*
embedding_size
,
sizeof
(
T
)
*
embedding_size
);
in_indexs
[
shard_id
]
+=
1
;
}
}
}
};
}
// namespace operators
}
// namespace paddle
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