That is true. I went for a simple implementation of the layer norm and included it in the tensorli definition. But it would have been better to define it as a moduli for clarity.