chunked/nltk_chunking/model_doc_layoutlm/chunk_2.txt

It obtains state-of-the-art results
on several downstream tasks:

form understanding: the FUNSD dataset (a collection of 199 annotated
  forms comprising more than 30,000 words).