Architectures · intermediate · concept 90 of 311 · Updated Nov 4, 2026
Encoder-Decoder Models
An encoder-decoder model reads an input into an internal representation, then generates an output from it, one step at a time. The pattern fits any task that turns one sequence into another, such as translation or summarization. The original transformer used this design; GPT-style models keep only the decoder, which is a useful fact when reading any model's architecture.
Top AI contributors on this
- Noam ShazeerOpenAI
- Christopher ManningStanford University / AIX Ventures
- Aidan GomezCohere
- Ashish VaswaniNvidia (via Essential AI)
- Albert GuCarnegie Mellon University / Cartesia
Key terms
Sequence to sequenceEncoderDecoderCross-attentionLatent representation
Learn these first
Where you meet it in the real world
Machine translation, summarization, speech recognition, image captioning
Guides and articles
Sequence to sequence and attention ↗
Lena Voita
The encoder-decoder architecture ↗
Aston Zhang and others