The Llama model attention map with 3 documents is represented as follows: ./visualization-tools/vis.ipynb reproduces the visualization results in the paper. We provide more visualization tools under .