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
6b0c57cf
P
Paddle
项目概览
PaddlePaddle
/
Paddle
1 年多 前同步成功
通知
2302
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
未验证
提交
6b0c57cf
编写于
1月 20, 2022
作者:
Z
zhangbo9674
提交者:
GitHub
1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
Fix master weight bug for multi_tensor optimizer(momentum, adam) (#38991)
* fix mp * support merged_momentum for mp
上级
c0f27282
变更
3
显示空白变更内容
内联
并排
Showing
3 changed file
with
68 addition
and
60 deletion
+68
-60
paddle/fluid/operators/optimizers/merged_momentum_op.h
paddle/fluid/operators/optimizers/merged_momentum_op.h
+60
-50
python/paddle/optimizer/adam.py
python/paddle/optimizer/adam.py
+4
-5
python/paddle/optimizer/momentum.py
python/paddle/optimizer/momentum.py
+4
-5
未找到文件。
paddle/fluid/operators/optimizers/merged_momentum_op.h
浏览文件 @
6b0c57cf
...
@@ -48,13 +48,13 @@ struct MergedMomentumKernelParam
...
@@ -48,13 +48,13 @@ struct MergedMomentumKernelParam
T
*
PADDLE_RESTRICT
params
[
N
];
T
*
PADDLE_RESTRICT
params
[
N
];
const
T
*
PADDLE_RESTRICT
grads
[
N
];
const
T
*
PADDLE_RESTRICT
grads
[
N
];
MT
*
PADDLE_RESTRICT
velocitys
[
N
];
MT
*
PADDLE_RESTRICT
velocitys
[
N
];
const
M
T
*
PADDLE_RESTRICT
lr
;
const
M
ultiPrecisionType
<
MT
>
*
PADDLE_RESTRICT
lr
;
MT
mu
;
MT
mu
;
MT
rescale_grad
;
MT
rescale_grad
;
uint32_t
param_num
;
uint32_t
param_num
;
HOSTDEVICE
void
operator
()(
size_t
i
)
const
{
HOSTDEVICE
void
operator
()(
size_t
i
)
const
{
const
auto
lr_val
=
*
lr
;
const
MT
lr_val
=
static_cast
<
MT
>
(
*
lr
)
;
for
(
uint32_t
idx
=
0
;
idx
<
param_num
;
++
idx
)
{
for
(
uint32_t
idx
=
0
;
idx
<
param_num
;
++
idx
)
{
auto
size
=
sizes
[
idx
];
auto
size
=
sizes
[
idx
];
if
(
i
>=
size
)
continue
;
if
(
i
>=
size
)
continue
;
...
@@ -81,8 +81,22 @@ struct MergedMomentumKernelParam
...
@@ -81,8 +81,22 @@ struct MergedMomentumKernelParam
template
<
typename
DeviceContext
,
typename
T
>
template
<
typename
DeviceContext
,
typename
T
>
class
MergedMomentumOpKernel
:
public
framework
::
OpKernel
<
T
>
{
class
MergedMomentumOpKernel
:
public
framework
::
OpKernel
<
T
>
{
using
MPType
=
typename
operators
::
details
::
MPTypeTrait
<
T
>::
Type
;
public:
public:
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
void
Compute
(
const
framework
::
ExecutionContext
&
ctx
)
const
override
{
const
bool
multi_precision
=
ctx
.
Attr
<
bool
>
(
"multi_precision"
);
if
(
multi_precision
)
{
InnerCompute
<
MPType
>
(
ctx
,
multi_precision
);
}
else
{
InnerCompute
<
T
>
(
ctx
,
multi_precision
);
}
}
private:
template
<
typename
MT
>
void
InnerCompute
(
const
framework
::
ExecutionContext
&
ctx
,
const
bool
multi_precision
)
const
{
auto
params
=
ctx
.
MultiInput
<
framework
::
Tensor
>
(
"Param"
);
auto
params
=
ctx
.
MultiInput
<
framework
::
Tensor
>
(
"Param"
);
auto
params_out
=
ctx
.
MultiOutput
<
framework
::
Tensor
>
(
"ParamOut"
);
auto
params_out
=
ctx
.
MultiOutput
<
framework
::
Tensor
>
(
"ParamOut"
);
size_t
n
=
params
.
size
();
size_t
n
=
params
.
size
();
...
@@ -133,7 +147,6 @@ class MergedMomentumOpKernel : public framework::OpKernel<T> {
...
@@ -133,7 +147,6 @@ class MergedMomentumOpKernel : public framework::OpKernel<T> {
auto
master_params
=
ctx
.
MultiInput
<
framework
::
Tensor
>
(
"MasterParam"
);
auto
master_params
=
ctx
.
MultiInput
<
framework
::
Tensor
>
(
"MasterParam"
);
auto
master_params_out
=
auto
master_params_out
=
ctx
.
MultiOutput
<
framework
::
Tensor
>
(
"MasterParamOut"
);
ctx
.
MultiOutput
<
framework
::
Tensor
>
(
"MasterParamOut"
);
auto
multi_precision
=
ctx
.
Attr
<
bool
>
(
"multi_precision"
);
if
(
multi_precision
)
{
if
(
multi_precision
)
{
PADDLE_ENFORCE_EQ
(
PADDLE_ENFORCE_EQ
(
n
,
master_params
.
size
(),
n
,
master_params
.
size
(),
...
@@ -206,18 +219,16 @@ class MergedMomentumOpKernel : public framework::OpKernel<T> {
...
@@ -206,18 +219,16 @@ class MergedMomentumOpKernel : public framework::OpKernel<T> {
<<
", regularization_coeffs.size(): "
<<
", regularization_coeffs.size(): "
<<
regularization_coeffs
.
size
();
<<
regularization_coeffs
.
size
();
using
MPType
=
typename
operators
::
details
::
MPTypeTrait
<
T
>::
Type
;
auto
&
dev_ctx
=
ctx
.
template
device_context
<
DeviceContext
>();
auto
&
dev_ctx
=
ctx
.
template
device_context
<
DeviceContext
>();
if
(
lrs
.
size
()
==
1
&&
use_nesterov
==
false
&&
if
(
lrs
.
size
()
==
1
&&
use_nesterov
==
false
&&
regularization_methods
.
size
()
==
0
)
{
regularization_methods
.
size
()
==
0
)
{
#define PADDLE_LAUNCH_MERGED_MOMENTUM_KERNEL(kMultiPrecision) \
#define PADDLE_LAUNCH_MERGED_MOMENTUM_KERNEL(kMultiPrecision) \
MergedMomentumKernelParam<T, M
PType
, kMultiPrecision> kernel_params; \
MergedMomentumKernelParam<T, M
T
, kMultiPrecision> kernel_params; \
constexpr auto kMaxMergedNum = decltype(kernel_params)::N; \
constexpr auto kMaxMergedNum = decltype(kernel_params)::N; \
size_t kernel_num = (n + kMaxMergedNum - 1) / kMaxMergedNum; \
size_t kernel_num = (n + kMaxMergedNum - 1) / kMaxMergedNum; \
kernel_params.mu = static_cast<M
PType
>(mu); \
kernel_params.mu = static_cast<M
T
>(mu); \
kernel_params.rescale_grad = static_cast<M
PType
>(rescale_grad); \
kernel_params.rescale_grad = static_cast<M
T
>(rescale_grad); \
kernel_params.lr = lrs[0]->data<MPType>(); \
kernel_params.lr = lrs[0]->data<MPType>(); \
for (size_t i = 0; i < kernel_num; ++i) { \
for (size_t i = 0; i < kernel_num; ++i) { \
size_t start = i * kMaxMergedNum; \
size_t start = i * kMaxMergedNum; \
...
@@ -230,9 +241,9 @@ class MergedMomentumOpKernel : public framework::OpKernel<T> {
...
@@ -230,9 +241,9 @@ class MergedMomentumOpKernel : public framework::OpKernel<T> {
kernel_params.sizes[j] = size; \
kernel_params.sizes[j] = size; \
kernel_params.params[j] = params_out[j + start]->data<T>(); \
kernel_params.params[j] = params_out[j + start]->data<T>(); \
kernel_params.grads[j] = grads[j + start]->data<T>(); \
kernel_params.grads[j] = grads[j + start]->data<T>(); \
kernel_params.velocitys[j] = velocitys_out[j + start]->data<M
PType
>(); \
kernel_params.velocitys[j] = velocitys_out[j + start]->data<M
T
>(); \
kernel_params.SetMasterParam( \
kernel_params.SetMasterParam( \
j, kMultiPrecision ? master_params_out[j + start]->data<M
PType
>() \
j, kMultiPrecision ? master_params_out[j + start]->data<M
T
>() \
: nullptr); \
: nullptr); \
} \
} \
platform::ForRange<DeviceContext> for_range(dev_ctx, max_size); \
platform::ForRange<DeviceContext> for_range(dev_ctx, max_size); \
...
@@ -254,34 +265,33 @@ class MergedMomentumOpKernel : public framework::OpKernel<T> {
...
@@ -254,34 +265,33 @@ class MergedMomentumOpKernel : public framework::OpKernel<T> {
?
RegularizationType
::
kL2DECAY
?
RegularizationType
::
kL2DECAY
:
RegularizationType
::
kNONE
;
:
RegularizationType
::
kNONE
;
M
PType
regularization_coeff
=
static_cast
<
MPType
>
(
0.0
);
M
T
regularization_coeff
=
static_cast
<
MT
>
(
0.0
);
if
(
regularization_coeffs
.
size
()
!=
0
)
{
if
(
regularization_coeffs
.
size
()
!=
0
)
{
regularization_coeff
=
regularization_coeff
=
static_cast
<
MT
>
(
regularization_coeffs
[
idx
]);
static_cast
<
MPType
>
(
regularization_coeffs
[
idx
]);
}
}
auto
lr_temp
=
lrs
.
size
()
>
1
?
lrs
[
idx
]
:
lrs
[
0
];
auto
lr_temp
=
lrs
.
size
()
>
1
?
lrs
[
idx
]
:
lrs
[
0
];
const
M
PType
*
master_in_data
=
const
M
T
*
master_in_data
=
multi_precision
?
master_params
[
idx
]
->
data
<
M
PType
>
()
:
nullptr
;
multi_precision
?
master_params
[
idx
]
->
data
<
M
T
>
()
:
nullptr
;
M
PType
*
master_out_data
=
M
T
*
master_out_data
=
multi_precision
?
master_params_out
[
idx
]
->
data
<
M
PType
>
()
:
nullptr
;
multi_precision
?
master_params_out
[
idx
]
->
data
<
M
T
>
()
:
nullptr
;
if
(
platform
::
is_cpu_place
(
ctx
.
GetPlace
()))
{
if
(
platform
::
is_cpu_place
(
ctx
.
GetPlace
()))
{
CPUDenseMomentumFunctor
<
M
PType
>
functor
;
CPUDenseMomentumFunctor
<
M
T
>
functor
;
functor
(
params
[
idx
],
grads
[
idx
],
velocitys
[
idx
],
lr_temp
,
mu
,
functor
(
params
[
idx
],
grads
[
idx
],
velocitys
[
idx
],
lr_temp
,
use_nesterov
,
regularization_flag
,
regularization_coeff
,
static_cast
<
MT
>
(
mu
),
use_nesterov
,
regularization_flag
,
params_out
[
idx
],
velocitys_out
[
idx
]);
regularization_coeff
,
params_out
[
idx
],
velocitys_out
[
idx
]);
VLOG
(
10
)
<<
"Launch MergedMomentum cpu kernel."
;
VLOG
(
10
)
<<
"Launch MergedMomentum cpu kernel."
;
}
else
if
(
platform
::
is_gpu_place
(
ctx
.
GetPlace
()))
{
}
else
if
(
platform
::
is_gpu_place
(
ctx
.
GetPlace
()))
{
platform
::
ForRange
<
DeviceContext
>
for_range
(
platform
::
ForRange
<
DeviceContext
>
for_range
(
static_cast
<
const
DeviceContext
&>
(
ctx
.
device_context
()),
static_cast
<
const
DeviceContext
&>
(
ctx
.
device_context
()),
params
[
idx
]
->
numel
());
params
[
idx
]
->
numel
());
#define PADDLE_LAUNCH_DENSE_MTMOMENTUM_KERNEL(__nesterov, __reg_type) \
#define PADDLE_LAUNCH_DENSE_MTMOMENTUM_KERNEL(__nesterov, __reg_type) \
DenseMomentumFunctor<T, M
PType, __reg_type, __nesterov> functor(
\
DenseMomentumFunctor<T, M
T, __reg_type, __nesterov> functor(
\
params[idx]->data<T>(), grads[idx]->data<T>(), \
params[idx]->data<T>(), grads[idx]->data<T>(), \
velocitys[idx]->data<M
PType>(), lr_temp->data<MPType>(), master_in_data,
\
velocitys[idx]->data<M
T>(), lr_temp->data<MPType>(), master_in_data,
\
mu, rescale_grad, params[idx]->numel(), regularization_coeff,
\
static_cast<MT>(mu), static_cast<MT>(rescale_grad),
\
params
_out[idx]->data<T>(), velocitys_out[idx]->data<MPType>(),
\
params
[idx]->numel(), regularization_coeff, params_out[idx]->data<T>(),
\
master_out_data);
\
velocitys_out[idx]->data<MT>(), master_out_data);
\
for_range(functor);
for_range(functor);
if
(
use_nesterov
)
{
if
(
use_nesterov
)
{
if
(
regularization_flag
==
RegularizationType
::
kL2DECAY
)
{
if
(
regularization_flag
==
RegularizationType
::
kL2DECAY
)
{
...
...
python/paddle/optimizer/adam.py
浏览文件 @
6b0c57cf
...
@@ -551,8 +551,7 @@ class Adam(Optimizer):
...
@@ -551,8 +551,7 @@ class Adam(Optimizer):
multi_tensor_list
=
[
'FP32_LODTensor'
,
'FP16_LODTensor'
]
multi_tensor_list
=
[
'FP32_LODTensor'
,
'FP16_LODTensor'
]
for
key
in
multi_tensor_list
:
for
key
in
multi_tensor_list
:
if
len
(
self
.
_param_dict
[
key
])
>
0
:
if
len
(
self
.
_param_dict
[
key
])
>
0
:
if
key
==
'FP32_LODTensor'
:
find_master
=
self
.
_multi_precision
and
key
==
'FP16_LODTensor'
self
.
_multi_precision
=
False
_beta1
=
self
.
_beta1
if
not
isinstance
(
_beta1
=
self
.
_beta1
if
not
isinstance
(
self
.
_beta1
,
Variable
)
else
self
.
_beta1
.
numpy
().
item
(
0
)
self
.
_beta1
,
Variable
)
else
self
.
_beta1
.
numpy
().
item
(
0
)
...
@@ -571,7 +570,7 @@ class Adam(Optimizer):
...
@@ -571,7 +570,7 @@ class Adam(Optimizer):
self
.
_beta2_pow_acc_dict
[
key
],
self
.
_beta2_pow_acc_dict
[
key
],
self
.
_master_weight_dict
[
key
],
'epsilon'
,
self
.
_epsilon
,
self
.
_master_weight_dict
[
key
],
'epsilon'
,
self
.
_epsilon
,
'beta1'
,
_beta1
,
'beta2'
,
_beta2
,
'multi_precision'
,
'beta1'
,
_beta1
,
'beta2'
,
_beta2
,
'multi_precision'
,
self
.
_multi_precision
)
find_master
)
else
:
else
:
inputs
=
{
inputs
=
{
"Param"
:
self
.
_param_dict
[
key
],
"Param"
:
self
.
_param_dict
[
key
],
...
@@ -594,11 +593,11 @@ class Adam(Optimizer):
...
@@ -594,11 +593,11 @@ class Adam(Optimizer):
"beta1"
:
_beta1
,
"beta1"
:
_beta1
,
"beta2"
:
_beta2
"beta2"
:
_beta2
}
}
if
self
.
_multi_precision
:
if
find_master
:
inputs
[
"MasterParam"
]
=
self
.
_master_weight_dict
[
key
]
inputs
[
"MasterParam"
]
=
self
.
_master_weight_dict
[
key
]
outputs
[
"MasterParamOut"
]
=
self
.
_master_weight_dict
[
outputs
[
"MasterParamOut"
]
=
self
.
_master_weight_dict
[
key
]
key
]
attrs
[
"multi_precision"
]
=
self
.
_multi_precision
attrs
[
"multi_precision"
]
=
find_master
target_block
.
append_op
(
target_block
.
append_op
(
type
=
"merged_adam"
,
type
=
"merged_adam"
,
inputs
=
inputs
,
inputs
=
inputs
,
...
...
python/paddle/optimizer/momentum.py
浏览文件 @
6b0c57cf
...
@@ -464,8 +464,7 @@ class Momentum(Optimizer):
...
@@ -464,8 +464,7 @@ class Momentum(Optimizer):
multi_tensor_list
=
[
'FP32_LODTensor'
,
'FP16_LODTensor'
]
multi_tensor_list
=
[
'FP32_LODTensor'
,
'FP16_LODTensor'
]
for
key
in
multi_tensor_list
:
for
key
in
multi_tensor_list
:
if
len
(
self
.
_param_dict
[
key
])
>
0
:
if
len
(
self
.
_param_dict
[
key
])
>
0
:
if
key
==
'FP32_LODTensor'
:
find_master
=
self
.
_multi_precision
and
key
==
'FP16_LODTensor'
self
.
_multi_precision
=
False
if
framework
.
in_dygraph_mode
():
if
framework
.
in_dygraph_mode
():
_
,
_
,
_
=
_C_ops
.
merged_momentum
(
_
,
_
,
_
=
_C_ops
.
merged_momentum
(
...
@@ -478,7 +477,7 @@ class Momentum(Optimizer):
...
@@ -478,7 +477,7 @@ class Momentum(Optimizer):
self
.
_regularization_method_dict
[
key
],
self
.
_regularization_method_dict
[
key
],
'regularization_coeff'
,
'regularization_coeff'
,
self
.
_regularization_coeff_dict
[
key
],
'multi_precision'
,
self
.
_regularization_coeff_dict
[
key
],
'multi_precision'
,
self
.
_multi_precision
)
find_master
)
else
:
else
:
inputs
=
{
inputs
=
{
"Param"
:
self
.
_param_dict
[
key
],
"Param"
:
self
.
_param_dict
[
key
],
...
@@ -498,11 +497,11 @@ class Momentum(Optimizer):
...
@@ -498,11 +497,11 @@ class Momentum(Optimizer):
"regularization_coeff"
:
"regularization_coeff"
:
self
.
_regularization_coeff_dict
[
key
],
self
.
_regularization_coeff_dict
[
key
],
}
}
if
self
.
_multi_precision
:
if
find_master
:
inputs
[
"MasterParam"
]
=
self
.
_master_weight_dict
[
key
]
inputs
[
"MasterParam"
]
=
self
.
_master_weight_dict
[
key
]
outputs
[
"MasterParamOut"
]
=
self
.
_master_weight_dict
[
outputs
[
"MasterParamOut"
]
=
self
.
_master_weight_dict
[
key
]
key
]
attrs
[
"multi_precision"
]
=
self
.
_multi_precision
attrs
[
"multi_precision"
]
=
find_master
target_block
.
append_op
(
target_block
.
append_op
(
type
=
"merged_momentum"
,
type
=
"merged_momentum"
,
inputs
=
inputs
,
inputs
=
inputs
,
...
...
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