Prediction Length Issue

[ikhsansdqq]

Hi @lucidrains,

I've noticed that you recreated the iTransformer based on the thuml/iTransformer library. I have been comparing your library and Thuml's library and it's completely different usage, especially on the pred_length parameter. I noticed in your package if I don't declaring the pred_length it will not work. Since pred_length is important and required in order to run the script, should I rely on the pred_length output because the model will automatically generate a dict output or should I create a custom training loop first and the model inside that custom training loop? (e.g. using PyTorch / Lightning)