Model is not learning: problem with custom activation function and/or with custom loss function

Question

class QuaternionLoss(torch.nn.Module):
    def __init__(self):
        super(QuaternionLoss, self).__init__()
        
    def forward(self, output, target):
        los