chunked/nltk_chunking/_attention/chunk_2.txt

Longformer and reformer are models that try to be more efficient and
use a sparse version of the attention matrix to speed up training.