chunked/nltk_chunking/model_doc_segformer/chunk_21.txt

Next, they throw
  away the classification head, and replace it by the all-MLP decode head.