Glossary term
Glossary term
Architecture
Transformer architecture combining an encoder for input understanding and a decoder for output generation, suited for seq2seq tasks.
T5 (Google) is trained as an encoder-decoder model on a text-to-text framework - used by Google Search for question answering, summarisation, and translation across multiple product surfaces.
FLAN-T5 (Google) fine-tunes T5 on hundreds of instruction-following tasks - deployed by Hugging Face users for zero-shot classification, summarisation, and structured data extraction without task-specific training.
mBART (Meta) is an encoder-decoder model for multilingual machine translation - used in production for low-resource language pairs where decoder-only models lack sufficient training signal.