What is the difference between NeuralPDE and Nvidia Modulus sym in python?

I want to know what is the difference between the default nn architecture in Julia and Nvidia Modulus Sym. I mean here we have Lux and create network for each variable. what I mean about defuault in julia is something like the PDE system examples in NeuralPDE.jl like Linear parabolic system of PDEs · NeuralPDE.jl . In nvidia modulus is almost the same but they mention that the default number layers is 6 and number of nerouns is 512. when I solve the same problem in both, 12 Gb GPU memory is getting involved while in Julia 40 Gb GPU memory. I think there is a difference in their architecture and definitely am doing or thinking something mistake.