This project is a time series forecasting model using the Temporal Fusion Transformer (TFT) deep learning architecture. The model is trained and evaluated on the M4 competition dataset, achieving state-of-the-art results in multi-step forecasting tasks.