Comment by manux

Yeah the "net" in GFlowNet refers to how the underlying state is interpreted, not to an architecture. It is a way to train generative models, on any kind of discrete data (like graphs).

Source: am first author of original GFlowNet paper.