The [[transformer model]]'s decoder is responsible for generating an output sequence based on the encoded input sequence. The decoder operates in an autoregressive manner, generating one token at a time while considering both the previously generated tokens and the encoded input information. The transformer model's decoder is primarily used for tasks that involve sequence generation, such as language translation, text summarization, and text generation. The decoder part of a transformer includes several key components: [[masked multi-head self-attention]], [[encoder-decoder attention]], [[positional encoding]], [[feedforward neural network]], [[transformer output layer]]. During sequence generation, the decoder generates tokens step by step, using the previously generated tokens and the encoded input information to inform its decisions. At each step, the model selects the token with the highest probability as the predicted token, and this predicted token is then fed as input to the next step. The decoder is a crucial part of the transformer architecture for tasks that involve generating sequences, as it enables the model to learn complex relationships between input and output sequences and generate coherent and contextually appropriate outputs. It's important to note that while the encoder and decoder are separate parts of the transformer model, they often share parameters and are trained jointly to optimize the end-to-end performance of the model on the given task. [[transformer encoder]] < [[Hands-on LLMs]]/[[2 LLMs and Transformers]] > [[masked multi-head self-attention]]