A PyTorch experiment demonstrating the core lookup capability of the Transformer architecture. By isolating Cross-Attention, this project generates a clean heatmap visualization that shows a perfect, targeted information retrieval from the Encoder's knowledge base.