chunked/nltk_chunking/model_doc_perceiver/chunk_16.txt

In this way, the time and memory requirements don't
depend on the length of the inputs anymore, as one uses a fixed amount of latent variables, like 256 or 512.