chunked/nltk_chunking/model_doc_perceiver/chunk_15.txt

Perceiver aims to solve this issue by, instead of performing self-attention on the inputs, perform it on a set
of latent variables, and only use the inputs for cross-attention.