chunked/nltk_chunking/_model_summary/chunk_19.txt

Attention is only computed within a local window, and the window is shifted between attention layers to create connections to help the model learn better.