Tests

Neural networks are tricky to train and with something like gperc where the entirity of code is just writing big config files (not YAML 😛), testing of underlying layer is very crucial. To that extent I have written tests for:

1. test_cifar10_forward: to test if it can process CIFAR10 dataset

2. test_image_overfit: to test if we overfit the model on a small random dataset, can it’s loss go all the way to zero.