What is the purpose of a Seq2Seq transformer?

• A. Encodes input then directly outputs a classification.
• B. Produces an output image.
• C. Processes input with an encoder and generates an output sequence using a decoder.
• D. Only decodes inputs without encoding.

Answer: (C)

Seq2Seq transformers are designed to map one sequence to another by using an encoder to read the input and a decoder to generate the output sequence. The encoder processes the entire input and creates a rich representation that captures its meaning and structure. The decoder then produces the target sequence one token at a time, attending to the encoder’s representations and to what it has already generated. This setup is ideal for tasks where the output is another sequence — like translating a sentence or summarizing text — because the output depends on the full input context, not just a single label or an image. Decoding without encoding would lack input context, and simply outputting a classification would compress the input to a single decision rather than a sequence. Generating an image involves a different kind of model, so it isn’t what a Seq2Seq transformer is designed to do.