This isn’t really supported, since if you convert it you will lose the indices. As a work around, you can either access the underlying data (xx.data Note: this is an internal feature, so it may break at any point!).
A better approach is to just use the JuMPArray as much as possible. If you want to write it to CSV, you can always build an array manually.
I, J, K = axes(xx)
[xx[i,j,k] for i in I, j in J, k in K]
Good luck using JuMP!
p.s. in the future, it’s easier to help if you follow