Look how far we've come! Now we add random .unsqueeze(-1) and .permute(-1,-2) until our PyTorch models run without shape errors!