p83571920
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

!21217 modify loss scale manager example 

Merge pull request !21217 from wangnan39/code_docs_loss_scale
This commit is contained in:
i-robot 2021-08-02 11:19:13 +00:00 committed by Gitee
parent 3ff3e6f57a 8aa0081bd5
commit 7f71a99993
6 changed files with 17 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
mindspore/nn/cell.py
View File

@ -1403,8 +1403,7 @@ class GraphCell(Cell):
Examples:
>>> import numpy as np
>>> import mindspore.nn as nn
>>> from mindspore import Tensor
>>> from mindspore.train import export, load
>>> from mindspore import Tensor, export, load
>>>
>>> net = nn.Conv2d(1, 1, kernel_size=3, weight_init="ones")
>>> input = Tensor(np.ones([1, 1, 3, 3]).astype(np.float32))

25
mindspore/nn/wrap/loss_scale.py
View File

@ -88,15 +88,14 @@ class DynamicLossScaleUpdateCell(Cell):
Examples:
>>> import numpy as np
>>> from mindspore import Tensor, Parameter, nn
>>> from mindspore.ops import operations as P
>>> from mindspore.nn.wrap.cell_wrapper import WithLossCell
>>> import mindspore.ops as ops
>>>
>>> class Net(nn.Cell):
... def __init__(self, in_features, out_features):
... super(Net, self).__init__()
... self.weight = Parameter(Tensor(np.ones([in_features, out_features]).astype(np.float32)),
... name='weight')
... self.matmul = P.MatMul()
... self.matmul = ops.MatMul()
...
... def construct(self, x):
... output = self.matmul(x, self.weight)
@ -106,7 +105,7 @@ class DynamicLossScaleUpdateCell(Cell):
>>> net = Net(in_features, out_features)
>>> loss = nn.MSELoss()
>>> optimizer = nn.Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
>>> net_with_loss = WithLossCell(net, loss)
>>> net_with_loss = nn.WithLossCell(net, loss)
>>> manager = nn.DynamicLossScaleUpdateCell(loss_scale_value=2**12, scale_factor=2, scale_window=1000)
>>> train_network = nn.TrainOneStepWithLossScaleCell(net_with_loss, optimizer, scale_sense=manager)
>>> input = Tensor(np.ones([out_features, in_features]), mindspore.float32)
@ -179,15 +178,14 @@ class FixedLossScaleUpdateCell(Cell):
Examples:
>>> import numpy as np
>>> from mindspore import Tensor, Parameter, nn
>>> from mindspore.ops import operations as P
>>> from mindspore.nn.wrap.cell_wrapper import WithLossCell
>>> from mindspore.ops as ops
>>>
>>> class Net(nn.Cell):
... def __init__(self, in_features, out_features):
... super(Net, self).__init__()
... self.weight = Parameter(Tensor(np.ones([in_features, out_features]).astype(np.float32)),
... name='weight')
... self.matmul = P.MatMul()
... self.matmul = ops.MatMul()
...
... def construct(self, x):
... output = self.matmul(x, self.weight)
@ -197,7 +195,7 @@ class FixedLossScaleUpdateCell(Cell):
>>> net = Net(in_features, out_features)
>>> loss = nn.MSELoss()
>>> optimizer = nn.Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
>>> net_with_loss = WithLossCell(net, loss)
>>> net_with_loss = nn.WithLossCell(net, loss)
>>> manager = nn.FixedLossScaleUpdateCell(loss_scale_value=2**12)
>>> train_network = nn.TrainOneStepWithLossScaleCell(net_with_loss, optimizer, scale_sense=manager)
>>> input = Tensor(np.ones([out_features, in_features]), mindspore.float32)
@ -253,16 +251,15 @@ class TrainOneStepWithLossScaleCell(TrainOneStepCell):
Examples:
>>> import numpy as np
>>> from mindspore import Tensor, Parameter, nn
>>> from mindspore.ops import operations as P
>>> from mindspore.nn.wrap.cell_wrapper import WithLossCell
>>> from mindspore.common import dtype as mstype
>>> from mindspore.ops as ops
>>> from mindspore import dtype as mstype
>>>
>>> class Net(nn.Cell):
... def __init__(self, in_features, out_features):
... super(Net, self).__init__()
... self.weight = Parameter(Tensor(np.ones([in_features, out_features]).astype(np.float32)),
... name='weight')
... self.matmul = P.MatMul()
... self.matmul = ops.MatMul()
...
... def construct(self, x):
... output = self.matmul(x, self.weight)
@ -273,7 +270,7 @@ class TrainOneStepWithLossScaleCell(TrainOneStepCell):
>>> net = Net(in_features, out_features)
>>> loss = nn.MSELoss()
>>> optimizer = nn.Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
>>> net_with_loss = WithLossCell(net, loss)
>>> net_with_loss = nn.WithLossCell(net, loss)
>>> manager = nn.DynamicLossScaleUpdateCell(loss_scale_value=2**12, scale_factor=2, scale_window=1000)
>>> train_network = nn.TrainOneStepWithLossScaleCell(net_with_loss, optimizer, scale_sense=manager)
>>> input = Tensor(np.ones([out_features, in_features]), mindspore.float32)
@ -284,7 +281,7 @@ class TrainOneStepWithLossScaleCell(TrainOneStepCell):
>>> net = Net(in_features, out_features)
>>> loss = nn.MSELoss()
>>> optimizer = nn.Momentum(net.trainable_params(), learning_rate=0.1, momentum=0.9)
>>> net_with_loss = WithLossCell(net, loss)
>>> net_with_loss = nn.WithLossCell(net, loss)
>>> inputs = Tensor(np.ones([size, in_features]).astype(np.float32))
>>> label = Tensor(np.zeros([size, out_features]).astype(np.float32))
>>> scaling_sens = Tensor(np.full((1), np.finfo(np.float32).max), dtype=mstype.float32)

3
mindspore/train/loss_scale_manager.py
View File

@ -115,8 +115,7 @@ class DynamicLossScaleManager(LossScaleManager):
scale_window (int): Maximum continuous normal steps when there is no overflow. Default: 2000.
Examples:
>>> from mindspore import Model, nn
>>> from mindspore.train.loss_scale_manager import DynamicLossScaleManager
>>> from mindspore import Model, nn, DynamicLossScaleManager
>>>
>>> net = Net()
>>> loss_scale_manager = DynamicLossScaleManager()

6
mindspore/train/model.py
View File

@ -615,8 +615,7 @@ class Model:
Default: -1.
Examples:
>>> from mindspore import Model, nn
>>> from mindspore.train.loss_scale_manager import FixedLossScaleManager
>>> from mindspore import Model, nn, FixedLossScaleManager
>>>
>>> # For details about how to build the dataset, please refer to the tutorial
>>> # document on the official website.
@ -872,10 +871,9 @@ class Model:
>>> # mindspore.cn.
>>> import numpy as np
>>> import mindspore as ms
>>> from mindspore import Model, context, Tensor, nn
>>> from mindspore import Model, context, Tensor, nn, FixedLossScaleManager
>>> from mindspore.context import ParallelMode
>>> from mindspore.communication import init
>>> from mindspore.train.loss_scale_manager import FixedLossScaleManager
>>>
>>> context.set_context(mode=context.GRAPH_MODE)
>>> init()

3
mindspore/train/serialization.py
View File

@ -331,8 +331,7 @@ def load(file_name, **kwargs):
Examples:
>>> import numpy as np
>>> import mindspore.nn as nn
>>> from mindspore import Tensor
>>> from mindspore.train import export, load
>>> from mindspore import Tensor, export, load
>>>
>>> net = nn.Conv2d(1, 1, kernel_size=3, weight_init="ones")
>>> input = Tensor(np.ones([1, 1, 3, 3]).astype(np.float32))

2
mindspore/train/train_thor/convert_utils.py
View File

@ -195,7 +195,7 @@ class ConvertModelUtils():
Examples:
>>> from mindspore.nn.optim import thor
>>> from mindspore.train.model import Model
>>> from mindspore.train.loss_scale_manager import FixedLossScaleManager
>>> from mindspore import FixedLossScaleManager
>>>
>>> net = Net()
>>> loss_manager = FixedLossScaleManager(128, drop_overflow_update=False)