How to get the beta_{i,j} for each residual branch?

[SefaZeng]

Figure 7 in your paper plots the weights of each residual branch for post-LN and pre-LN. I am wondering how to get these weights exactly. I note that there is a parameter called plot_variance, looks like it outputs sqrt(Var[a_j]). Then how can I get the beta_{i, j}?

Thx for advancing!