提交 c34c3e5f 编写于 作者: X Xiaoda 提交者: Gitee

update tutorials/source_en/advanced_use/mixed_precision.md.

上级 7ad03657
...@@ -36,7 +36,6 @@ This document describes the computation process by using examples of automatic a ...@@ -36,7 +36,6 @@ This document describes the computation process by using examples of automatic a
## Automatic Mixed Precision ## Automatic Mixed Precision
To use the automatic mixed precision, you need to invoke the corresponding API, which takes the network to be trained and the optimizer as the input. This API converts the operators of the entire network into FP16 operators (except the BatchNorm and Loss operators). To use the automatic mixed precision, you need to invoke the corresponding API, which takes the network to be trained and the optimizer as the input. This API converts the operators of the entire network into FP16 operators (except the BatchNorm and Loss operators).
In addition, after the mixed precision is employed, the loss scale must be used to avoid data overflow.
The procedure is as follows: The procedure is as follows:
1. Introduce the MindSpore mixed precision API. 1. Introduce the MindSpore mixed precision API.
...@@ -49,57 +48,44 @@ A code example is as follows: ...@@ -49,57 +48,44 @@ A code example is as follows:
```python ```python
import numpy as np import numpy as np
import mindspore.nn as nn import mindspore.nn as nn
import mindspore.common.dtype as mstype
from mindspore import Tensor, context from mindspore import Tensor, context
from mindspore.ops import operations as P from mindspore.ops import operations as P
from mindspore.nn import WithLossCell
from mindspore.nn import Momentum from mindspore.nn import Momentum
from mindspore.nn.loss import MSELoss
# The interface of Auto_mixed precision # The interface of Auto_mixed precision
from mindspore.train import amp from mindspore import amp
context.set_context(mode=context.GRAPH_MODE) context.set_context(mode=context.GRAPH_MODE)
context.set_context(device_target="Ascend") context.set_context(device_target="Ascend")
# Define network # Define network
class LeNet5(nn.Cell): class Net(nn.Cell):
def __init__(self): def __init__(self, input_channel, out_channel):
super(LeNet5, self).__init__() super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 6, 5, pad_mode='valid') self.dense = nn.Dense(input_channel, out_channel)
self.conv2 = nn.Conv2d(6, 16, 5, pad_mode='valid') self.relu = P.ReLU()
self.fc1 = nn.Dense(16 * 5 * 5, 120)
self.fc2 = nn.Dense(120, 84)
self.fc3 = nn.Dense(84, 10)
self.relu = nn.ReLU()
self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
self.flatten = P.Flatten()
def construct(self, x): def construct(self, x):
x = self.max_pool2d(self.relu(self.conv1(x))) x = self.dense(x)
x = self.max_pool2d(self.relu(self.conv2(x))) x = self.relu(x)
x = self.flatten(x)
x = self.relu(self.fc1(x))
x = self.relu(self.fc2(x))
x = self.fc3(x)
return x return x
# Initialize network # Initialize network
net = LeNet5() net = Net(512, 128)
# Define training data, label and sens # Define training data, label
predict = Tensor(np.ones([1, 1, 32, 32]).astype(np.float32) * 0.01) predict = Tensor(np.ones([64, 512]).astype(np.float32) * 0.01)
label = Tensor(np.zeros([1, 10]).astype(np.float32)) label = Tensor(np.zeros([64, 128]).astype(np.float32))
scaling_sens = Tensor(np.full((1), 1.0), dtype=mstype.float32)
# Define Loss and Optimizer # Define Loss and Optimizer
loss = MSELoss() loss = nn.SoftmaxCrossEntropyWithLogits()
optimizer = Momentum(params=net.trainable_params(), learning_rate=0.1, momentum=0.9) optimizer = Momentum(params=net.trainable_params(), learning_rate=0.1, momentum=0.9)
net_with_loss = WithLossCell(net, loss) train_network = amp.build_train_network(net, optimizer, loss, level="O2", loss_scale_manager=None)
train_network = amp.build_train_network(net_with_loss, optimizer, level="O2")
# Run training # Run training
output = train_network(predict, label, scaling_sens) output = train_network(predict, label)
``` ```
...@@ -110,66 +96,53 @@ MindSpore also supports manual mixed precision. It is assumed that only one dens ...@@ -110,66 +96,53 @@ MindSpore also supports manual mixed precision. It is assumed that only one dens
The following is the procedure for implementing manual mixed precision: The following is the procedure for implementing manual mixed precision:
1. Define the network. This step is similar to step 2 in the automatic mixed precision. 1. Define the network. This step is similar to step 2 in the automatic mixed precision.
2. Configure the mixed precision. Use net.to_float(mstype.float16) to set all operators of the cell and its sub-cells to FP16. Then, configure the fc3 to FP32. 2. Configure the mixed precision. Use net.to_float(mstype.float16) to set all operators of the cell and its sub-cells to FP16. Then, configure the dense to FP32.
3. Use TrainOneStepWithLossScaleCell to encapsulate the network model and optimizer. 3. Use TrainOneStepCell to encapsulate the network model and optimizer.
A code example is as follows: A code example is as follows:
```python ```python
import numpy as np import numpy as np
import mindspore.nn as nn import mindspore.nn as nn
import mindspore.common.dtype as mstype import mindspore.common.dtype as mstype
from mindspore import Tensor, context from mindspore import Tensor, context
from mindspore.ops import operations as P from mindspore.ops import operations as P
from mindspore.nn import WithLossCell, TrainOneStepWithLossScaleCell from mindspore.nn import WithLossCell, TrainOneStepCell
from mindspore.nn import Momentum from mindspore.nn import Momentum
from mindspore.nn.loss import MSELoss
context.set_context(mode=context.GRAPH_MODE) context.set_context(mode=context.GRAPH_MODE)
context.set_context(device_target="Ascend") context.set_context(device_target="Ascend")
# Define network # Define network
class LeNet5(nn.Cell): class Net(nn.Cell):
def __init__(self): def __init__(self, input_channel, out_channel):
super(LeNet5, self).__init__() super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 6, 5, pad_mode='valid') self.dense = nn.Dense(input_channel, out_channel)
self.conv2 = nn.Conv2d(6, 16, 5, pad_mode='valid') self.relu = P.ReLU()
self.fc1 = nn.Dense(16 * 5 * 5, 120)
self.fc2 = nn.Dense(120, 84)
self.fc3 = nn.Dense(84, 10)
self.relu = nn.ReLU()
self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
self.flatten = P.Flatten()
def construct(self, x): def construct(self, x):
x = self.max_pool2d(self.relu(self.conv1(x))) x = self.dense(x)
x = self.max_pool2d(self.relu(self.conv2(x))) x = self.relu(x)
x = self.flatten(x)
x = self.relu(self.fc1(x))
x = self.relu(self.fc2(x))
x = self.fc3(x)
return x return x
# Initialize network and set mixing precision # Initialize network and set mixing precision
net = LeNet5() net = Net(512, 128)
net.to_float(mstype.float16) net.to_float(mstype.float16)
net.fc3.to_float(mstype.float32) net.dense.to_float(mstype.float32)
# Define training data, label and sens # Define training data, label
predict = Tensor(np.ones([1, 1, 32, 32]).astype(np.float32) * 0.01) predict = Tensor(np.ones([64, 512]).astype(np.float32) * 0.01)
label = Tensor(np.zeros([1, 10]).astype(np.float32)) label = Tensor(np.zeros([64, 128]).astype(np.float32))
scaling_sens = Tensor(np.full((1), 1.0), dtype=mstype.float32)
# Define Loss and Optimizer # Define Loss and Optimizer
net.set_train() loss = nn.SoftmaxCrossEntropyWithLogits()
loss = MSELoss()
optimizer = Momentum(params=net.trainable_params(), learning_rate=0.1, momentum=0.9) optimizer = Momentum(params=net.trainable_params(), learning_rate=0.1, momentum=0.9)
net_with_loss = WithLossCell(net, loss) net_with_loss = WithLossCell(net, loss)
train_network = TrainOneStepWithLossScaleCell(net_with_loss, optimizer) train_network = TrainOneStepCell(net_with_loss, optimizer)
train_network.set_train() train_network.set_train()
# Run training # Run training
output = train_network(predict, label, scaling_sens) output = train_network(predict, label)
``` ```
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