Another odd fact is that the models learn almost identical parameters.
Here are the parameters for the simple model:
Here are for the parallel model:
group_1 is the group in question.
group_1