Description

We need to implement “recurrent” transformers to do faster inference of many batches of short prediction horizon predictions. This model could be combined components from the TFT.