BART¶

Overview¶

The Bart model was proposed in BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension by Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Ves Stoyanov and Luke Zettlemoyer on 29 Oct, 2019.

According to the abstract,

Bart uses a standard seq2seq/machine translation architecture with a bidirectional encoder (like BERT) and a left-to-right decoder (like GPT).
The pretraining task involves randomly shuffling the order of the original sentences and a novel in-filling scheme, where spans of text are replaced with a single mask token.
BART is particularly effective when fine tuned for text generation but also works well for comprehension tasks. It matches the performance of RoBERTa with comparable training resources on GLUE and SQuAD, achieves new state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks, with gains of up to 6 ROUGE.

Examples and scripts for fine-tuning BART and other models for sequence to sequence tasks can be found in examples/pytorch/summarization/.
An example of how to train BartForConditionalGeneration with a Hugging Face datasets object can be found in this forum discussion.
Distilled checkpoints are described in this paper.

Bart doesn’t use token_type_ids for sequence classification. Use BartTokenizer or encode() to get the proper splitting.
The forward pass of BartModel will create the decoder_input_ids if they are not passed. This is different than some other modeling APIs. A typical use case of this feature is mask filling.
Model predictions are intended to be identical to the original implementation when force_bos_token_to_be_generated=True. This only works, however, if the string you pass to fairseq.encode() starts with a space.
generate() should be used for conditional generation tasks like summarization, see the example in that docstrings.