使用PyNative模式调试

使用PyNative模式调试

概述

MindSpore支持两种运行模式，在调试或者运行方面做了不同的优化:

PyNative模式：也称动态图模式，将神经网络中的各个算子逐一下发执行，方便用户编写和调试神经网络模型。
Graph模式：也称静态图模式或者图模式，将神经网络模型编译成一整张图，然后下发执行。该模式利用图优化等技术提高运行性能，同时有助于规模部署和跨平台运行。

默认情况下，MindSpore处于PyNative模式，可以通过context.set_context(mode=context.GRAPH_MODE)切换为Graph模式；同样地，MindSpore处于Graph模式时，可以通过 context.set_context(mode=context.PYNATIVE_MODE)切换为PyNative模式。

PyNative模式下，支持执行单算子、普通函数和网络，以及单独求梯度的操作。下面将详细介绍使用方法和注意事项。

执行单算子

执行单个算子，并打印相关结果，如下例所示。

Copyimport numpy as np
import mindspore.nn as nn
from mindspore import context, Tensor
 
context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
 
conv = nn.Conv2d(3, 4, 3, bias_init='zeros')
input_data = Tensor(np.ones([1, 3, 5, 5]).astype(np.float32))
output = conv(input_data)
print(output.asnumpy())

输出：

Copy[[[[-0.02190447 -0.05208071 -0.05208071 -0.05208071 -0.06265172]
[-0.01529094 -0.05286242 -0.05286242 -0.05286242 -0.04228776]
[-0.01529094 -0.05286242 -0.05286242 -0.05286242 -0.04228776]
[-0.01529094 -0.05286242 -0.05286242 -0.05286242 -0.04228776]
[-0.01430791 -0.04892948 -0.04892948 -0.04892948 -0.01096004]]
 
[[ 0.00802889 -0.00229866 -0.00229866 -0.00229866 -0.00471579]
[ 0.01172971 0.02172665 0.02172665 0.02172665 0.03261888]
[ 0.01172971 0.02172665 0.02172665 0.02172665 0.03261888]
[ 0.01172971 0.02172665 0.02172665 0.02172665 0.03261888]
[ 0.01784375 0.01185635 0.01185635 0.01185635 0.01839031]]
 
[[ 0.04841832 0.03321705 0.03321705 0.03321705 0.0342317 ]
[ 0.0651359 0.04310361 0.04310361 0.04310361 0.03355784]
[ 0.0651359 0.04310361 0.04310361 0.04310361 0.03355784]
[ 0.0651359 0.04310361 0.04310361 0.04310361 0.03355784]
[ 0.04680437 0.03465693 0.03465693 0.03465693 0.00171057]]
 
[[-0.01783456 -0.00459451 -0.00459451 -0.00459451 0.02316688]
[ 0.01295831 0.00879035 0.00879035 0.00879035 0.01178642]
[ 0.01295831 0.00879035 0.00879035 0.00879035 0.01178642]
[ 0.01295831 0.00879035 0.00879035 0.00879035 0.01178642]
[ 0.05016355 0.03958241 0.03958241 0.03958241 0.03443141]]]]

执行普通函数

将若干算子组合成一个函数，然后直接通过函数调用的方式执行这些算子，并打印相关结果，如下例所示。

示例代码

Copyimport numpy as np
from mindspore import context, Tensor
from mindspore.ops import functional as F
 
context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
 
def tensor_add_func(x, y):
    z = F.tensor_add(x, y)
    z = F.tensor_add(z, x)
    return z
 
x = Tensor(np.ones([3, 3], dtype=np.float32))
y = Tensor(np.ones([3, 3], dtype=np.float32))
output = tensor_add_func(x, y)
print(output.asnumpy())

输出

Copy[[3. 3. 3.]
 [3. 3. 3.]
 [3. 3. 3.]]

提升PyNative性能

为了提高PyNative模式下的前向计算任务执行速度，MindSpore提供了Staging功能，该功能可以在PyNative模式下将Python函数或者Python类的方法编译成计算图，通过图优化等技术提高运行速度，如下例所示。

Copyimport numpy as np
import numpy as np
import mindspore.nn as nn
from mindspore import context, Tensor
import mindspore.ops.operations as P
from mindspore.common.api import ms_function
 
context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
 
class TensorAddNet(nn.Cell):
    def __init__(self):
        super(TensorAddNet, self).__init__()
        self.add = P.TensorAdd()
 
    @ms_function
    def construct(self, x, y):
        res = self.add(x, y)
        return res
 
x = Tensor(np.ones([4, 4]).astype(np.float32))
y = Tensor(np.ones([4, 4]).astype(np.float32))
net = TensorAddNet()
 
z = net(x, y) # Staging mode
tensor_add = P.TensorAdd()
res = tensor_add(x, z) # PyNative mode
print(res.asnumpy())

输出

Copy[[3. 3. 3. 3.]
 [3. 3. 3. 3.]
 [3. 3. 3. 3.]
 [3. 3. 3. 3.]]

上述示例代码中，在TensorAddNet类的construct之前加装了ms_function装饰器，该装饰器会将construct方法编译成计算图，在给定输入之后，以图的形式下发执行，而上一示例代码中的F.tensor_add会直以普通的PyNative的方式执行。

需要说明的是，加装了ms_function装饰器的函数中，如果包含不需要进行参数训练的算子（如pooling、tensor_add等算子），则这些算子可以在被装饰的函数中直接调用，如下例所示。

示例代码

Copyimport numpy as np
import mindspore.nn as nn
from mindspore import context, Tensor
import mindspore.ops.operations as P
from mindspore.common.api import ms_function
 
context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
 
tensor_add = P.TensorAdd()
 
@ms_function
def tensor_add_fn(x, y):
    res = tensor_add(x, y)
    return res
 
x = Tensor(np.ones([4, 4]).astype(np.float32))
y = Tensor(np.ones([4, 4]).astype(np.float32))
z = tensor_add_fn(x, y)
print(z.asnumpy())

输出

Copy[[2. 2. 2. 2.]
 [2. 2. 2. 2.]
 [2. 2. 2. 2.]
 [2. 2. 2. 2.]]

如果被装饰的函数中包含了需要进行参数训练的算子（如Convolution、BatchNorm等算子），则这些算子必须在被装饰等函数之外完成实例化操作，如下例所示。

示例代码

Copyimport numpy as np
import mindspore.nn as nn
from mindspore import context, Tensor
from mindspore.common.api import ms_function
 
context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
 
conv_obj = nn.Conv2d(in_channels=3, out_channels=4, kernel_size=3, stride=2, padding=0)
@ms_function
def conv_fn(x):
    res = conv_obj(x)
    return res
 
input_data = np.random.randn(2, 3, 6, 6).astype(np.float32)
z = conv_fn(Tensor(input_data))
print(z.asnumpy())

输出

Copy[[[[ 0.10377571 -0.0182163 -0.05221086]
[ 0.1428334 -0.01216263 0.03171652]
[-0.00673915 -0.01216291 0.02872104]]
 
[[ 0.02906547 -0.02333629 -0.0358406 ]
[ 0.03805163 -0.00589525 0.04790922]
[-0.01307234 -0.00916951 0.02396654]]
 
[[ 0.01477884 -0.06549098 -0.01571796]
[ 0.00526886 -0.09617482 0.04676902]
[-0.02132788 -0.04203424 0.04523344]]
 
[[ 0.04590619 -0.00251453 -0.00782715]
[ 0.06099087 -0.03445276 0.00022781]
[ 0.0563223 -0.04832596 -0.00948266]]]
 
[[[ 0.08444098 -0.05898955 -0.039262 ]
[ 0.08322686 -0.0074796 0.0411371 ]
[-0.02319113 0.02128408 -0.01493311]]
 
[[ 0.02473745 -0.02558945 -0.0337843 ]
[-0.03617039 -0.05027632 -0.04603915]
[ 0.03672804 0.00507637 -0.08433761]]
 
[[ 0.09628943 0.01895323 -0.02196114]
[ 0.04779419 -0.0871575 0.0055248 ]
[-0.04382382 -0.00511185 -0.01168541]]
 
[[ 0.0534859 0.02526264 0.04755395]
[-0.03438103 -0.05877855 0.06530266]
[ 0.0377498 -0.06117418 0.00546303]]]]

调试网络训练模型

PyNative模式下，还可以支持单独求梯度的操作。如下例所示，可通过grad_all求该函数或者网络所有的输入梯度。

示例代码

Copyfrom mindspore.ops import composite as C
import mindspore.context as context
 
context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
 
def mul(x, y):
    return x * y
 
def mainf(x, y):
    return C.grad_all(mul)(x, y)
 
print(mainf(1,2))

输出

Copy(2, 1)

在进行网络训练时，求得梯度然后调用优化器对参数进行优化（暂不支持在反向计算梯度的过程中设置断点），然后再利用前向计算loss，从而实现在PyNative模式下进行网络训练。

完整LeNet示例代码

Copyimport numpy as np
import mindspore.nn as nn
import mindspore.ops.operations as P
from mindspore.nn import Dense
from mindspore import context, Tensor, ParameterTuple
from mindspore.common.initializer import TruncatedNormal
from mindspore.ops import composite as C
from mindspore.common import dtype as mstype
from mindspore.nn.wrap.cell_wrapper import WithLossCell
from mindspore.nn.loss import SoftmaxCrossEntropyWithLogits
from mindspore.nn.optim import Momentum
 
context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU")
 
def conv(in_channels, out_channels, kernel_size, stride=1, padding=0):
    """weight initial for conv layer"""
    weight = weight_variable()
    return nn.Conv2d(in_channels, out_channels,
                     kernel_size=kernel_size, stride=stride, padding=padding,
                     weight_init=weight, has_bias=False, pad_mode="valid")
 
def fc_with_initialize(input_channels, out_channels):
    """weight initial for fc layer"""
    weight = weight_variable()
    bias = weight_variable()
    return nn.Dense(input_channels, out_channels, weight, bias)
 
def weight_variable():
    """weight initial"""
    return TruncatedNormal(0.02)
 
class LeNet5(nn.Cell):
    """
    Lenet network
    Args:
        num_class (int): Num classes. Default: 10.
 
    Returns:
        Tensor, output tensor
 
    Examples:
        >>> LeNet(num_class=10)
    """
    def __init__(self, num_class=10):
        super(LeNet5, self).__init__()
        self.num_class = num_class
        self.batch_size = 32
        self.conv1 = conv(1, 6, 5)
        self.conv2 = conv(6, 16, 5)
        self.fc1 = fc_with_initialize(16 * 5 * 5, 120)
        self.fc2 = fc_with_initialize(120, 84)
        self.fc3 = fc_with_initialize(84, self.num_class)
        self.relu = nn.ReLU()
        self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
        self.reshape = P.Reshape()
 
    def construct(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.max_pool2d(x)
        x = self.conv2(x)
        x = self.relu(x)
        x = self.max_pool2d(x)
        x = self.reshape(x, (self.batch_size, -1))
        x = self.fc1(x)
        x = self.relu(x)
        x = self.fc2(x)
        x = self.relu(x)
        x = self.fc3(x)
        return x
 
 
class GradWrap(nn.Cell):
    """ GradWrap definition """
    def __init__(self, network):
        super(GradWrap, self).__init__(auto_prefix=False)
        self.network = network
        self.weights = ParameterTuple(filter(lambda x: x.requires_grad, network.get_parameters()))
 
    def construct(self, x, label):
        weights = self.weights
        return C.grad_by_list(self.network, weights)(x, label)
 
net = LeNet5()
optimizer = Momentum(filter(lambda x: x.requires_grad, net.get_parameters()), 0.1, 0.9)
criterion = nn.SoftmaxCrossEntropyWithLogits(is_grad=False, sparse=True)
net_with_criterion = WithLossCell(net, criterion)
train_network = GradWrap(net_with_criterion)
train_network.set_train()
 
input_data = Tensor(np.ones([net.batch_size, 1, 32, 32]).astype(np.float32) * 0.01)
label = Tensor(np.ones([net.batch_size]).astype(np.int32))
output = net(Tensor(input_data))
loss_output = criterion(output, label)
grads = train_network(input_data, label)
success = optimizer(grads)
loss = loss_output.asnumpy()
print(loss)

输出

Copy2.3050091

上述执行方式中，可以在construct函数任意需要的地方设置断点，获取网络执行的中间结果，通过pdb的方式对网络进行调试。