What is the difference between encoder-only, decoder-only, and encoder-decoder architectures?

Encoder-only models understand context, decoder-only generate text, and encoder-decoder models perform sequence-to-sequence tasks.

Encoder-only models like BERT focus on bidirectional understanding. Decoder-only models like GPT generate autoregressive text. Encoder-decoder models like T5 combine both for translation and summarization. Each architecture differs in attention masking, training objective, and downstream suitability.