Encoder
Also known as: Source Encoder
In an encoder–decoder system, the encoder reads the source sequence (for example the English sentence in translation) and builds representations of it.
In early neural MT the encoder was usually an RNN that walked the source left to right and ended with one final hidden state. Modern encoder–decoder Transformers instead produce a vector (or stack of vectors) for every source position.
The encoder’s job is understanding the input; it does not emit the target tokens. That generation is left to the decoder.