未验证 提交 cc2d8b36 编写于 作者: C ceci3 提交者: GitHub

Add convert_syncbn (#2526)

* add docs
上级 0432a859
......@@ -62,3 +62,27 @@ SyncBatchNorm
print(hidden1.numpy())
# [[[[0.26824948, 1.0936325],[0.26824948, -1.6301316]],[[ 0.8095662, -0.665287],[-1.2744656, 1.1301866 ]]]]
方法
:::::::::
convert_sync_batchnorm(layer)
'''''''''''''''''''''''''''''
该接口用于把 ``BatchNorm*d`` 层转换为 ``SyncBatchNorm`` 层。
参数:
- **layer** (paddle.nn.Layer) - 包含一个或多个 ``BatchNorm*d`` 层的模型。
返回:
如果原始模型中有 ``BatchNorm*d`` 层, 则把 ``BatchNorm*d`` 层转换为 ``SyncBatchNorm`` 层的原始模型。
**代码示例**
.. code-block:: python
import paddle
import paddle.nn as nn
paddle.disable_static()
model = nn.Sequential(nn.Conv2d(3, 5, 3), nn.BatchNorm2d(5))
sync_model = nn.SyncBatchNorm.convert_sync_batchnorm(model)
......@@ -62,3 +62,26 @@ SyncBatchNorm
print(hidden1.numpy())
# [[[[0.26824948, 1.0936325],[0.26824948, -1.6301316]],[[ 0.8095662, -0.665287],[-1.2744656, 1.1301866 ]]]]
方法
:::::::::
convert_sync_batchnorm(layer)
'''''''''''''''''''''''''''''
该接口用于把 ``BatchNorm*d`` 层转换为 ``SyncBatchNorm`` 层。
参数:
- **layer** (paddle.nn.Layer) - 包含一个或多个 ``BatchNorm*d`` 层的模型。
返回:
如果原始模型中有 ``BatchNorm*d`` 层, 则把 ``BatchNorm*d`` 层转换为 ``SyncBatchNorm`` 层的原始模型。
**代码示例**
.. code-block:: python
import paddle
import paddle.nn as nn
paddle.disable_static()
model = nn.Sequential(nn.Conv2d(3, 5, 3), nn.BatchNorm2d(5))
sync_model = nn.SyncBatchNorm.convert_sync_batchnorm(model)
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册