A transformer-based sequence-to-sequence model that predicts Formula 1 lap pace deltas using tyre compound, stint age, fuel load, and micro-sector telemetry data. This project uses multivariate time-series forecasting with attention mechanisms while providing actionable insights for F1 broadcasters, teams, and strategy analysts. The pipeline flows from Fast-F1 API data collection through feature extraction to a transformer architecture built with PyTorch Lightning, deployed on an NVIDIA RTX A4500 GPU cluster using SLURM scheduling. The model outperforms traditional baselines (exponential smoothing, linear regression, random forest) and includes explainability features through SHAP values and attention maps. GitHub