INPUT OUTPUT ENCODER DECODER yₜ₋₁ c
The input sequence is processed into a single fixed-size context vector.