CIFAR-10 and IMDB share a dataset split for validation and testing

[hazdzz]

I discovered that both CIFAR-10 and IMDB datasets utilize an identical dataset split for their validation and testing purposes. The relevant source code is presented below:

data_val = torch.load('./cifar10_test_vocab.pt').to(torch.int64)
labels_val = torch.load('./cifar10_test_targets_vocab.pt').to(torch.int64)

data_test = torch.load('./cifar10_test_vocab.pt').to(torch.int64)
labels_test = torch.load('./cifar10_test_targets_vocab.pt').to(torch.int64)

data_val = torch.load('./data/IMDB_test.pt').to(torch.int64)
labels_val = torch.load('./data/IMDB_test_targets.pt').to(torch.int64)

data_test = torch.load('./data/IMDB_test.pt').to(torch.int64)
labels_test = torch.load('./data/IMDB_test_targets.pt').to(torch.int64)