chunked/nltk_chunking/tasks_summarization/chunk_20.txt

It's more efficient to dynamically pad the sentences to the longest length in a batch during collation, instead of padding the whole dataset to the maximum length.