Hey guys im working on an R-GAN in python. I use pytorch and i have a problem. I use LSTM for my GAN to create about 240 time steps with an input size of 8(loc vetkor). I generate a Multivariate Gaussian Distribution (MVGD) for each time step. The input and the corresponding MVGD changes for each time step. With 20 time steps i get good generated distributions, but the higher i go it get's worse. The loss function of my generator diverges fast. Any ideas how i can improve my GAN ?