T5 (Text-to-Text)

← Back to the article

T5 treats many language jobs as “text in, text out.”

What it does

T5 treats many language jobs as “text in, text out.”

Unified: Doing many jobs with one system • Transfer: Moving skills

How it works

  • Take written text as the input.
  • Decide what kind of text answer is needed.
  • Write new text as the output.

Uses and applications

  • It can turn a long article into a short summary.
  • It can translate text into another language.
  • It can answer a question using text.

In short

T5 turns many language tasks into the same text-to-text shape.

Related methods

  • Transformer: The encoder-decoder architecture that T5 uses.
  • GPT: Generates continuations instead of framing every task as text in, text out.
  • BERT: Uses only the encoder side, mainly for understanding tasks.

Back to Language and text models · All topics

The interactive demonstration and the Python walkthrough for this topic need JavaScript.