chunked/nltk_chunking/model_doc_data2vec/chunk_5.txt

The core idea is to predict latent representations of the full input data based on a
masked view of the input in a selfdistillation setup using a standard Transformer architecture.