nn.Parameter() not learning

Why is self.A = nn.Parameter(F.normalize(torch.randn(d_model, state_size), p=2, dim=-1)) not learning ?