AI Glossary

Encoder-decoder model

Also called sequence-to-sequence model, seq2seq

What does “Encoder-decoder model” mean?

Encoder-decoder model: This design pairs an encoder, which reads and represents the full input, with a decoder, which generates the output while consulting that representation. It was the layout of the original transformer and fits tasks that map one sequence to another.

Encoder-decoder model visual guide: This design pairs an encoder, which reads and represents the full input, with a decoder, which generates the output while consulting that representation. It was the layout of the original transformer and fits tasks that map one sequence to another.
Visual guide Open full size (new tab)

What is an example of “Encoder-decoder model”?

Translation and summarization are classic uses; Google's T5 is a well-known encoder-decoder model.

Keep learning

See also

EncoderTermDecoderTermTransformerTerm

Same topic

Frontier modelTermGenerative adversarial networkTermEmbedding modelTermGPTTermDiffusion modelTerm