Did you manage to increase the performance? I also just started working with model based RL and was shocked how slow it was