Which statement is true about T5 and BART?

• A. They are examples of Seq2Seq transformers.
• B. They are encoder-only models.
• C. They are decoder-only models.
• D. They are image-processing models.

Answer: (A)

Both T5 and BART use an encoder-decoder design, which makes them sequence-to-sequence transformers. In this setup, the encoder reads the input text and creates a representation of it, and the decoder generates the output text conditioned on that representation. T5 treats every NLP task as a text-to-text problem, so whatever the task—translation, summarization, question answering—you input text and receive text as output. BART is trained as a denoising autoencoder, again with an encoder processing a corrupted input and a decoder reconstructing the original text, reinforcing the input-to-output mapping. They aren’t encoder-only models like BERT, nor decoder-only models like GPT, and they aren’t image-processing models. Hence, the statement that they are Seq2Seq transformers is the accurate one.