{"id":9606,"date":"2025-07-10T12:38:31","date_gmt":"2025-07-10T10:38:31","guid":{"rendered":"https:\/\/www.kubicek.ai\/?post_type=lexicon&#038;p=9606"},"modified":"2026-07-29T16:07:54","modified_gmt":"2026-07-29T14:07:54","slug":"bart","status":"publish","type":"lexicon","link":"https:\/\/www.kubicek.ai\/en\/lexicon\/bart\/","title":{"rendered":"BART"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><strong>Today we take a closer look at a model called BART.<\/strong> Think of it as a very clever language student who learns by repairing deliberately damaged texts. Give it a sentence with words missing or shuffled around and it learns to reconstruct it into its original, correct form. This training makes it a master both of understanding context and of producing new, meaningful text. That is why it is excellent for automatically writing summaries of long articles, or for translation between languages.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">Technical definition<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>BART<\/strong> is an acronym for <em>Bidirectional and Auto-Regressive Transformers<\/em>. It is a sequence-to-sequence (seq2seq) model based on the transformer architecture, introduced by the Meta AI research team in 2019. Its distinctive feature is the combination of two different approaches to pre-training. It uses a bidirectional encoder, much like the <a href=\"https:\/\/www.kubicek.ai\/en\/lexicon\/bert\/\" class=\"lex-link\">BERT<\/a> model, and an autoregressive decoder, characteristic of models such as GPT. This hybrid architecture allows BART to excel across a wide spectrum of natural language processing (NLP) tasks.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Pre-training of the model takes the form of denoising autoencoding. The input text is first deliberately corrupted by various methods such as masking, deleting or transposing tokens (words or parts of them) and filling in missing sections of text. The model&#8217;s goal is then to reconstruct the original, intact text from this corrupted input. This process teaches the model a deep understanding of grammar, semantics and contextual relationships in text.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Thanks to its encoder, which processes information from the whole sentence at once (bidirectionally), BART can grasp context effectively. Its autoregressive decoder then generates text sequentially, token by token, which is ideal for generative tasks. After the pre-training phase the model is further fine-tuned on specific datasets for particular applications. Typical uses include abstractive text summarisation, machine translation, question answering and text generation. In summarisation, for instance, the encoder processes the whole source document and the decoder then generates a concise, coherent summary of it.<\/p>\n","protected":false},"featured_media":0,"template":"","class_list":["post-9606","lexicon","type-lexicon","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon\/9606","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon"}],"about":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/types\/lexicon"}],"wp:attachment":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/media?parent=9606"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}