generating synthetic timeseries data with GAN models

Question

I'm developing a GAN model to generate synthetic energy timeseries data, based in this code https://git.opendfki.de/koochali/forgan/-/blob/master/lorenz_example.ipynb I'm beginner at GAN and a little bit confused about the data preparation. For GRU and LSTM models, I used to turn the data into lags (sequenaces) before passing it to the models. for example: the input ship will be:

sequance_lendth = 25
prediction_length = 3
input_features = 2
X_train = (batch_size, sequance_length, input_features)
y_train = (batch_size, prediction_length)

am I supposed to do the same in this case with GAN model ? The other wondering, in case I have multivariate time series (Price, stock index ...) and I want only to generate the price. Should I pass all features to the model?

This what I have done, but I'm not sure if this is reasonable or not.

class Generator(nn.Module):
    def __init__(self, input_features, output_dim, noise_size, x_batch_size, generator_latent_size):
        super().__init__()

        self.noise_size = noise_size
        self.x_batch_size = x_batch_size
        self.generator_latent_size = generator_latent_size
        self.cond_to_latent = nn.GRU(input_size=input_features,
                                     hidden_size=generator_latent_size)

        self.model = nn.Sequential(
            nn.Linear(in_features=generator_latent_size + noise_size,
                      out_features=generator_latent_size + noise_size),
            nn.ReLU(),
            nn.Linear(in_features=generator_latent_size + noise_size, out_features=prediction_length)
        )