CIFAR10 Experiment for Group Normalization

12import torch.nn as nn
13
14from labml import experiment
15from labml.configs import option
16from labml_helpers.module import Module
17from labml_nn.experiments.cifar10 import CIFAR10Configs
18from labml_nn.normalization.group_norm import GroupNorm

VGG model for CIFAR-10 classification

21class Model(Module):
26    def __init__(self, groups: int = 32):
27        super().__init__()
28        layers = []

RGB channels

30        in_channels = 3

Number of channels in each layer in each block

32        for block in [[64, 64], [128, 128], [256, 256, 256], [512, 512, 512], [512, 512, 512]]:

Convolution, Normalization and Activation layers

34            for channels in block:
35                layers += [nn.Conv2d(in_channels, channels, kernel_size=3, padding=1),
36                           GroupNorm(groups, channels),
37                           nn.ReLU(inplace=True)]
38                in_channels = channels

Max pooling at end of each block

40            layers += [nn.MaxPool2d(kernel_size=2, stride=2)]

Create a sequential model with the layers

43        self.layers = nn.Sequential(*layers)

Final logits layer

45        self.fc = nn.Linear(512, 10)
47    def forward(self, x):

The VGG layers

49        x = self.layers(x)

Reshape for classification layer

51        x = x.view(x.shape[0], -1)

Final linear layer

53        return self.fc(x)
56class Configs(CIFAR10Configs):

Number of groups

58    groups: int = 16

Create model

61@option(Configs.model)
62def model(c: Configs):
66    return Model(c.groups).to(c.device)
69def main():

Create experiment

71    experiment.create(name='cifar10', comment='group norm')

Create configurations

73    conf = Configs()

Load configurations

75    experiment.configs(conf, {
76        'optimizer.optimizer': 'Adam',
77        'optimizer.learning_rate': 2.5e-4,
78    })

Start the experiment and run the training loop

80    with experiment.start():
81        conf.run()

85if __name__ == '__main__':
86    main()