On a nonhierarchical version of the generalized random energy model