How to use Tensorflow WindowGenerator to make time series predictions, after model has already been trained

Question

I am following this official Tensorflow tutorial for time series for multi-variate, multi-step time series.

I understand that we split the time series dataset in training, validation, and test sets, and then used the WindowGenerator class to generate the batches of time series datasets, and then train several models.

However, once the model is trained, I am confused about the next step of using this model in practice. My dataset is a daily timeseries, and every day, I want to display a prediction based on the most recent (as of yesterday) time window (let's say last 2 months of data), and use is to make predictions for the next month. But I cannot figure out how to go from having the latest raw dataset in hand to using WindowGenerator class to give me back a single time window (yesterday through 2 months ago) which I would then use to predict the values for next month.

The WindowGenerator class is designed to work with train, val and test...but how to use it after the model is trained, when I am not interested in train, val, but only in test.