Created by: DrRyanHuang

• 环境 AIstudio平台 python3.7 PaddlePaddle 1.7.1

• 报错信息：直接运行会使得AIstudio重新启动，但是略微改动可以运行，后面会提供复现步骤

• 报错代码以及复现步骤：(已将代码简化)

from paddle.fluid.dygraph import Conv2D, Layer, Pool2D, Linear, Sequential, to_variable
from paddle.fluid.layers import flatten, concat

import paddle.fluid as fluid
import numpy as np

class Inception(Layer):
    
    def __init__(self, num_channels, ch1x1, ch3x3reduced, ch3x3, ch5x5reduced, ch5x5, pool_proj):
        
        super(Inception, self).__init__()

        self.branch11 = Conv2D(num_channels=3,
                            num_filters=10,
                            filter_size=3,
                            stride=1,
                            padding=1,
                            act='leaky_relu',
                            bias_attr=False,
                            # param_attr=conv_weight,
         )

    def forward(self, inputs):

        print(inputs.shape)
        print(self.branch11)
        branch1 = self.branch11(inputs)
        print(branch1.shape)

        return inputs

model = Inception(3, 5, 5, 5, 5, 5, 5)

with fluid.dygraph.guard():

    a = np.ones(shape=(4, 3, 5, 5), dtype=np.float32)
    a = to_variable(a)
    out = model(a)
    print(out.shape)

直接运行以上代码会使得AIstudio重新启动

将forward函数改一下让其报错

    def forward(self, inputs):

        print(inputs.shape)
        print(self.branch1)     # <---------------将 `branch11` 改为 `branch1`
        branch1 = self.branch11(inputs)
        print(branch1.shape)

        return inputs

按 Enter + shift运行，当然会报错说没有该属性:

AttributeError: 'Inception' object has no attribute 'branch1'

这时再将 forward 函数改回来：

    def forward(self, inputs):

        print(inputs.shape)
        print(self.branch11)     # <---------------将 `branch1` 改为 `branch11`
        branch1 = self.branch11(inputs)
        print(branch1.shape)

        return inputs

这时平台可运行，且没有问题 输出为：

[4, 3, 5, 5]
<paddle.fluid.dygraph.nn.Conv2D object at 0x7f8a69ae4170>
[4, 10, 5, 5]
[4, 3, 5, 5]

以上是我遇到的问题，请您帮忙看看，不胜感激！