I am working with monthly time series data that has only two columns: "month" and "value to predict." However, the data is non$-$stationary, and I want to leverage Transformers for accurate predictions.

What are the best preprocessing techniques to handle non$-$stationary time series data?How can I prepare this data for input into a Transformer model?How should I design the Transformer to predict future values, such as forecasting the next $6$ or $12$ months?

I would appreciate code snippets or step$-$by$-$step instructions to preprocess the data, create windows, train the Transformer, and perform future predictions. Thank you!