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
acbb5dbe
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
acbb5dbe
编写于
4月 28, 2022
作者:
R
ronnywang
提交者:
GitHub
4月 28, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
[CustomDevice] add amp support (#42035)
上级
c7a258fe
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
11 addition
and
4 deletion
+11
-4
paddle/fluid/imperative/amp_auto_cast.cc
paddle/fluid/imperative/amp_auto_cast.cc
+1
-0
python/paddle/fluid/dygraph/amp/auto_cast.py
python/paddle/fluid/dygraph/amp/auto_cast.py
+7
-2
python/paddle/fluid/dygraph/amp/loss_scaler.py
python/paddle/fluid/dygraph/amp/loss_scaler.py
+3
-2
未找到文件。
paddle/fluid/imperative/amp_auto_cast.cc
浏览文件 @
acbb5dbe
...
...
@@ -220,6 +220,7 @@ inline bool NeedCast(const std::shared_ptr<VarType>& var) {
paddle
::
platform
::
is_cuda_pinned_place
(
place
)
||
paddle
::
platform
::
is_xpu_place
(
place
)
||
paddle
::
platform
::
is_mlu_place
(
place
)
||
paddle
::
platform
::
is_custom_place
(
place
)
||
paddle
::
platform
::
is_npu_place
(
place
)
||
paddle
::
platform
::
is_npu_pinned_place
(
place
))
{
// CudaPinndePlace is added for varbase created by dataloader
...
...
python/paddle/fluid/dygraph/amp/auto_cast.py
浏览文件 @
acbb5dbe
...
...
@@ -276,9 +276,10 @@ def amp_guard(enable=True,
if
enable
and
not
(
tracer
.
_expected_place
.
is_gpu_place
()
or
tracer
.
_expected_place
.
is_xpu_place
()
or
tracer
.
_expected_place
.
is_mlu_place
()
or
tracer
.
_expected_place
.
is_npu_place
()):
tracer
.
_expected_place
.
is_npu_place
()
or
tracer
.
_expected_place
.
is_custom_place
()):
warnings
.
warn
(
'amp_guard can only be enabled on CUDAPlace, XPUPlace, MLUPlace,
and NPU
Place, current place is %s, so it makes no effect.'
'amp_guard can only be enabled on CUDAPlace, XPUPlace, MLUPlace,
NPUPlace, and Custom
Place, current place is %s, so it makes no effect.'
%
tracer
.
_expected_place
)
enable
=
False
# For npu:
...
...
@@ -293,6 +294,10 @@ def amp_guard(enable=True,
if
tracer
.
_expected_place
.
is_mlu_place
()
and
(
dtype
==
'bfloat16'
):
warnings
.
warn
(
'MLUPlace only support float16 amp.'
)
enable
=
False
# For custom device:
if
tracer
.
_expected_place
.
is_custom_place
()
and
(
dtype
==
'bfloat16'
):
warnings
.
warn
(
'CustomPlace only support float16 amp.'
)
enable
=
False
# For gpu float16: Compute Capability should >= 7.
# For gpu bfloat16: Compute Capability should >= 8 & CUDA Version should >= 11.
if
tracer
.
_expected_place
.
is_gpu_place
():
...
...
python/paddle/fluid/dygraph/amp/loss_scaler.py
浏览文件 @
acbb5dbe
...
...
@@ -107,9 +107,10 @@ class AmpScaler(object):
if
enable
and
not
(
tracer
.
_expected_place
.
is_gpu_place
()
or
tracer
.
_expected_place
.
is_xpu_place
()
or
tracer
.
_expected_place
.
is_mlu_place
()
or
tracer
.
_expected_place
.
is_npu_place
()):
tracer
.
_expected_place
.
is_npu_place
()
or
tracer
.
_expected_place
.
is_custom_place
()):
warnings
.
warn
(
'AmpScaler can only be enabled on CUDAPlace, XPUPlace, MLUPlace
and NPU
Place, current place is %s, so it makes no effect.'
'AmpScaler can only be enabled on CUDAPlace, XPUPlace, MLUPlace
, NPUPlace and Custom
Place, current place is %s, so it makes no effect.'
%
tracer
.
_expected_place
)
enable
=
False
...
...
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