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
models
提交
d0299b81
M
models
项目概览
PaddlePaddle
/
models
大约 2 年 前同步成功
通知
232
Star
6828
Fork
2962
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
602
列表
看板
标记
里程碑
合并请求
255
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
M
models
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
602
Issue
602
列表
看板
标记
里程碑
合并请求
255
合并请求
255
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
d0299b81
编写于
10月 22, 2019
作者:
Y
Yibing Liu
提交者:
GitHub
10月 22, 2019
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Fix div zero error in lr decay (#3694)
* Fix div zero error in lr decay * Update readme
上级
8db0319c
变更
5
隐藏空白更改
内联
并排
Showing
5 changed file
with
25 addition
and
12 deletion
+25
-12
PaddleNLP/language_representations_kit/BERT/README.md
PaddleNLP/language_representations_kit/BERT/README.md
+1
-1
PaddleNLP/language_representations_kit/BERT/model/classifier.py
...NLP/language_representations_kit/BERT/model/classifier.py
+2
-2
PaddleNLP/language_representations_kit/BERT/optimization.py
PaddleNLP/language_representations_kit/BERT/optimization.py
+18
-5
PaddleNLP/language_representations_kit/BERT/run_squad.py
PaddleNLP/language_representations_kit/BERT/run_squad.py
+3
-3
PaddleNLP/language_representations_kit/BERT/train.py
PaddleNLP/language_representations_kit/BERT/train.py
+1
-1
未找到文件。
PaddleNLP/language_representations_kit/BERT/README.md
浏览文件 @
d0299b81
...
@@ -70,7 +70,7 @@
...
@@ -70,7 +70,7 @@
```
```
## 安装
## 安装
本项目依赖于 Paddle Fluid
**1.
5.1
**
及以上版本,请参考
[
安装指南
](
http://www.paddlepaddle.org/#quick-start
)
进行安装。如果需要进行 TensorFlow 模型到 Paddle Fluid 参数的转换,则需要同时安装 TensorFlow 1.12。
本项目依赖于 Paddle Fluid
**1.
6.0
**
及以上版本,请参考
[
安装指南
](
http://www.paddlepaddle.org/#quick-start
)
进行安装。如果需要进行 TensorFlow 模型到 Paddle Fluid 参数的转换,则需要同时安装 TensorFlow 1.12。
## 预训练
## 预训练
...
...
PaddleNLP/language_representations_kit/BERT/model/classifier.py
浏览文件 @
d0299b81
...
@@ -26,13 +26,13 @@ def create_model(args, bert_config, num_labels, is_prediction=False):
...
@@ -26,13 +26,13 @@ def create_model(args, bert_config, num_labels, is_prediction=False):
input_fields
=
{
input_fields
=
{
'names'
:
[
'src_ids'
,
'pos_ids'
,
'sent_ids'
,
'input_mask'
,
'labels'
],
'names'
:
[
'src_ids'
,
'pos_ids'
,
'sent_ids'
,
'input_mask'
,
'labels'
],
'shapes'
:
[[
None
,
None
],
[
None
,
None
],
[
None
,
None
],
'shapes'
:
[[
None
,
None
],
[
None
,
None
],
[
None
,
None
],
[
-
1
,
args
.
max_seq_len
,
1
],
[
-
1
,
1
]],
[
None
,
args
.
max_seq_len
,
1
],
[
None
,
1
]],
'dtypes'
:
[
'int64'
,
'int64'
,
'int64'
,
'float32'
,
'int64'
],
'dtypes'
:
[
'int64'
,
'int64'
,
'int64'
,
'float32'
,
'int64'
],
'lod_levels'
:
[
0
,
0
,
0
,
0
,
0
],
'lod_levels'
:
[
0
,
0
,
0
,
0
,
0
],
}
}
inputs
=
[
inputs
=
[
fluid
.
layers
.
data
(
fluid
.
data
(
name
=
input_fields
[
'names'
][
i
],
name
=
input_fields
[
'names'
][
i
],
shape
=
input_fields
[
'shapes'
][
i
],
shape
=
input_fields
[
'shapes'
][
i
],
dtype
=
input_fields
[
'dtypes'
][
i
],
dtype
=
input_fields
[
'dtypes'
][
i
],
...
...
PaddleNLP/language_representations_kit/BERT/optimization.py
浏览文件 @
d0299b81
...
@@ -73,9 +73,10 @@ def optimization(loss,
...
@@ -73,9 +73,10 @@ def optimization(loss,
.
noam_decay
(
1
/
(
warmup_steps
*
(
learning_rate
**
2
)),
.
noam_decay
(
1
/
(
warmup_steps
*
(
learning_rate
**
2
)),
warmup_steps
)
warmup_steps
)
else
:
else
:
printf
(
print
(
"WARNING: noam decay should have postive warmup steps, using "
"WARNING: noam decay of learning rate should have postive warmup "
"constant learning rate instead!"
)
"steps but given {}, using constant learning rate instead!"
.
format
(
warmup_steps
))
scheduled_lr
=
fluid
.
layers
.
create_global_var
(
scheduled_lr
=
fluid
.
layers
.
create_global_var
(
name
=
fluid
.
unique_name
.
generate
(
"learning_rate"
),
name
=
fluid
.
unique_name
.
generate
(
"learning_rate"
),
shape
=
[
1
],
shape
=
[
1
],
...
@@ -83,8 +84,20 @@ def optimization(loss,
...
@@ -83,8 +84,20 @@ def optimization(loss,
dtype
=
'float32'
,
dtype
=
'float32'
,
persistable
=
True
)
persistable
=
True
)
elif
scheduler
==
'linear_warmup_decay'
:
elif
scheduler
==
'linear_warmup_decay'
:
scheduled_lr
=
linear_warmup_decay
(
learning_rate
,
warmup_steps
,
if
warmup_steps
>
0
:
num_train_steps
)
scheduled_lr
=
linear_warmup_decay
(
learning_rate
,
warmup_steps
,
num_train_steps
)
else
:
print
(
"WARNING: linear warmup decay of learning rate should have "
"postive warmup steps but given {}, use constant learning rate "
"instead!"
.
format
(
warmup_steps
))
scheduled_lr
=
fluid
.
layers
.
create_global_var
(
name
=
fluid
.
unique_name
.
generate
(
"learning_rate"
),
shape
=
[
1
],
value
=
learning_rate
,
dtype
=
'float32'
,
persistable
=
True
)
else
:
else
:
raise
ValueError
(
"Unkown learning rate scheduler, should be "
raise
ValueError
(
"Unkown learning rate scheduler, should be "
"'noam_decay' or 'linear_warmup_decay'"
)
"'noam_decay' or 'linear_warmup_decay'"
)
...
...
PaddleNLP/language_representations_kit/BERT/run_squad.py
浏览文件 @
d0299b81
...
@@ -111,7 +111,7 @@ def create_model(bert_config, is_training=False):
...
@@ -111,7 +111,7 @@ def create_model(bert_config, is_training=False):
input_fields
=
{
input_fields
=
{
'names'
:
[
'src_ids'
,
'pos_ids'
,
'sent_ids'
,
'input_mask'
,
'start_positions'
,
'end_positions'
],
'names'
:
[
'src_ids'
,
'pos_ids'
,
'sent_ids'
,
'input_mask'
,
'start_positions'
,
'end_positions'
],
'shapes'
:
[[
None
,
None
],
[
None
,
None
],
[
None
,
None
],
'shapes'
:
[[
None
,
None
],
[
None
,
None
],
[
None
,
None
],
[
-
1
,
args
.
max_seq_len
,
1
],
[
-
1
,
1
],
[
-
1
,
1
]],
[
None
,
args
.
max_seq_len
,
1
],
[
None
,
1
],
[
None
,
1
]],
'dtypes'
:
[
'dtypes'
:
[
'int64'
,
'int64'
,
'int64'
,
'float32'
,
'int64'
,
'int64'
],
'int64'
,
'int64'
,
'int64'
,
'float32'
,
'int64'
,
'int64'
],
'lod_levels'
:
[
0
,
0
,
0
,
0
,
0
,
0
],
'lod_levels'
:
[
0
,
0
,
0
,
0
,
0
,
0
],
...
@@ -120,13 +120,13 @@ def create_model(bert_config, is_training=False):
...
@@ -120,13 +120,13 @@ def create_model(bert_config, is_training=False):
input_fields
=
{
input_fields
=
{
'names'
:
[
'src_ids'
,
'pos_ids'
,
'sent_ids'
,
'input_mask'
,
'unique_id'
],
'names'
:
[
'src_ids'
,
'pos_ids'
,
'sent_ids'
,
'input_mask'
,
'unique_id'
],
'shapes'
:
[[
None
,
None
],
[
None
,
None
],
[
None
,
None
],
'shapes'
:
[[
None
,
None
],
[
None
,
None
],
[
None
,
None
],
[
-
1
,
args
.
max_seq_len
,
1
],
[
-
1
,
1
]],
[
None
,
args
.
max_seq_len
,
1
],
[
None
,
1
]],
'dtypes'
:
[
'dtypes'
:
[
'int64'
,
'int64'
,
'int64'
,
'float32'
,
'int64'
],
'int64'
,
'int64'
,
'int64'
,
'float32'
,
'int64'
],
'lod_levels'
:
[
0
,
0
,
0
,
0
,
0
],
'lod_levels'
:
[
0
,
0
,
0
,
0
,
0
],
}
}
inputs
=
[
fluid
.
layers
.
data
(
name
=
input_fields
[
'names'
][
i
],
inputs
=
[
fluid
.
data
(
name
=
input_fields
[
'names'
][
i
],
shape
=
input_fields
[
'shapes'
][
i
],
shape
=
input_fields
[
'shapes'
][
i
],
dtype
=
input_fields
[
'dtypes'
][
i
],
dtype
=
input_fields
[
'dtypes'
][
i
],
lod_level
=
input_fields
[
'lod_levels'
][
i
])
for
i
in
range
(
len
(
input_fields
[
'names'
]))]
lod_level
=
input_fields
[
'lod_levels'
][
i
])
for
i
in
range
(
len
(
input_fields
[
'names'
]))]
...
...
PaddleNLP/language_representations_kit/BERT/train.py
浏览文件 @
d0299b81
...
@@ -105,7 +105,7 @@ def create_model(bert_config):
...
@@ -105,7 +105,7 @@ def create_model(bert_config):
'lod_levels'
:
[
0
,
0
,
0
,
0
,
0
,
0
,
0
],
'lod_levels'
:
[
0
,
0
,
0
,
0
,
0
,
0
,
0
],
}
}
inputs
=
[
fluid
.
layers
.
data
(
name
=
input_fields
[
'names'
][
i
],
inputs
=
[
fluid
.
data
(
name
=
input_fields
[
'names'
][
i
],
shape
=
input_fields
[
'shapes'
][
i
],
shape
=
input_fields
[
'shapes'
][
i
],
dtype
=
input_fields
[
'dtypes'
][
i
],
dtype
=
input_fields
[
'dtypes'
][
i
],
lod_level
=
input_fields
[
'lod_levels'
][
i
])
for
i
in
range
(
len
(
input_fields
[
'names'
]))]
lod_level
=
input_fields
[
'lod_levels'
][
i
])
for
i
in
range
(
len
(
input_fields
[
'names'
]))]
...
...
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