Timeseries Forecasting with custom TFT Model

I would like to use the Keras API for time series forecasting with the Temporal Fusion Transformer. Other frameworks with a PyTorch backend have production-ready implementations (PyTorch Forecasting, DARTS). Keras is a very versatile and powerful multi-backend framework; thus, I decided to start the project: Keras-TFT with the powerful help of Antigravirty, as I am not a researcher nor a professional developer. I would love your feedback since it is just an attempt to replicate the TFT architecture. Any help or contribution is highly appreciated!