chunked/content_aware_chunking/_glossary/chunk_29.txt

During training, both BART and T5 will make the appropriate
  decoder_input_ids and decoder attention masks internally. They usually do not need to be supplied.