RNN based model types cannot run cross-validation properly

[egolub22]

When I try to run a cross-validation function where RNN based models are included in the NeuralForecast object. I get an error saying that RNN models cannot have a step size of over 1. I don't understand why this would be considering all of the cross-validation periods are within the dataframe and the stepsize merely determines the time stamp for the next window to begin. It should have no effect on an RNN models ability to get proper sequence data.

Bottom of the error: