chunked/nltk_chunking/model_doc_reformer/chunk_5.txt

The resulting model, the Reformer, performs on par with Transformer models
while being much more memory-efficient and much faster on long sequences.