An encoder is the component of a Transformer that turns an input sequence of Tokens into a stack of contextual vectors — its job is to 'understand', not to generate. Encoder-only models like BERT are particularly strong for classification, semantic search and Embedding generation, because every token can attend to context on both sides. In the classic Encoder-Decoder design the encoder reads the input and the decoder produces output conditioned on it; T5 and the original Transformer follow this pattern. Pure encoders have lost some spotlight in the generative-LLM era, but they remain quiet workhorses behind retrieval and embedding pipelines.
MEVZU N°124ISTANBULYEAR I — VOL. III
Glossary · Intermediate · 2017
Encoder
The Transformer component that turns input into a meaningful internal representation.
- EN — English term
- Encoder
- TR — Turkish term
- Kodlayıcı (Encoder)