Skip to content

MADLAD model

Jump to bottom
Isaac Schifferer edited this page Jul 18, 2025 · 1 revision

A few notes about working with the MADLAD model.

  • Unlike NLLB, you cannot specify a source language
  • Data format: Source sentences are prepended with the target language code in the form <2xx>, where xx is the language code. Nothing is added to target sentences.
  • MADLAD uses the same set of ISO codes as NLLB (meaning they follow the same standard, not that they were trained on the same set of languages)
Clone this wiki locally