Fix/Debug My TFT Model for Time Series Forecast

I'm seeking an expert in time series forecasting and TFT models to help me troubleshoot my current setup. I've already divided the project into three steps: data loading and preprocessing, training, and forecasting. Steps one and two have been executed and the training seems satisfactory. However, I'm encountering issues with step three. My forecasts are converging to a single line, with the forecast horizon of 35 days resulting in only one value or minimal variations. As such, the forecasts appear incorrect.

Currently, I am using Kedro as my pipeline, with three datasets located under /data. I previously loaded the data from Postgres and saved intermediate steps in a blob storage, but for the sake of this project, I have removed that logic so that you can work with it locally only.

The datasets I'm working with are purely time series data, and the deep learning framework in use for the TFT model is PyTorch. The specific issue I need assistance with is the forecast values not being correct.

Ideal skills and experience for the job include:
- Deep understanding of time series forecasting
- Proficient in using TFT models
- Extensive experience with PyTorch
- Familiarity with Kedro
- Ability to troubleshoot and correct forecasting errors

Please review my setup and let me know if you can assist.