15importtorch.nnasnn1617fromlabmlimportexperiment,logger18fromlabml.configsimportoption19fromlabml\_nn.experiments.cifar10importCIFAR10Configs,CIFAR10VGGModel20fromlabml\_nn.normalization.batch\_normimportBatchNorm

23classConfigs(CIFAR10Configs):

30pass

33classLargeModel(CIFAR10VGGModel):

40defconv\_block(self,in\_channels,out\_channels)-\>nn.Module:

44returnnn.Sequential(

46nn.Dropout(0.1),

48nn.Conv2d(in\_channels,out\_channels,kernel\_size=3,padding=1),

50BatchNorm(out\_channels,track\_running\_stats=False),

52nn.ReLU(inplace=True),53)

55def\_\_init\_\_(self):

57super().\_\_init\_\_([[64,64],[128,128],[256,256,256],[512,512,512],[512,512,512]])

60@option(Configs.model)61def\_large\_model(c:Configs):

65returnLargeModel().to(c.device)

68defmain():

70experiment.create(name='cifar10',comment='large model')

72conf=Configs()

74experiment.configs(conf,{75'optimizer.optimizer':'Adam',76'optimizer.learning\_rate':2.5e-4,77'is\_save\_models':True,78'epochs':20,79})

81experiment.add\_pytorch\_models({'model':conf.model})

83logger.inspect(params=(sum(p.numel()forpinconf.model.parameters()ifp.requires\_grad)))

85withexperiment.start():86conf.run()

90if\_\_name\_\_=='\_\_main\_\_':91main()