Weight decay norm

[cemde]

What does the optax.add_decayed_weights function actually implement?

For loss function $L$, penalty coefficient $\lambda$, and layer $l$, does it implement $L + \lambda \sum_l ||W_l||_2$ or $L +\lambda \sum_l ||W_l||_1$, i.e. L1 or L2 penalty? Whatever it implements is there a function for the other?