A basic sequence-to-sequence model that is composed of two recurrent neural networks (RNNs) connected by an attention mechanism. The encoder reads the input sequence and produces a fixed-length vector representation, while the decoder uses this representation to generate the output sequence.