Language model · Established · Advanced

T5

An encoder-decoder transformer that frames every NLP task as converting input text into output text.

What T5 is

T5's unified text-to-text formulation made a single architecture handle translation, summarisation, classification and question answering by changing only the prompt prefix.

How it works

Pretrained with a span-corruption objective, then fine-tuned per task. Instruction-tuned descendants extended it to many tasks at once.

Why it matters

It established the text-to-text framing that instruction-tuned models now take for granted.

Common uses

→Summarisation
→Translation
→Structured text transformation

Strengths

✓Clean task formulation
✓Strong for seq2seq work

Watch for

✓Superseded by larger decoder models for general use

Continue exploring

Transformer Instruction Tuning Machine Translation