XTREME-UP: A User-Centric Scarce-Data Benchmark for Under-Represented Languages
Download | Baselines | Evaluation | Public Results Tracker | Paper
This repository contains code to run experiments and evaluate models on XTREME-UP. It also contains the public results tracker with the results and predictions of all models that have been evaluated on XTREME-UP.
Download the Dataset
Once you've chosen which task to work on (above), you can download the data at the following URLs.
XTREME-UP data is primarily in:
Two exceptions:
- FLEURS audio data is available from
https://storage.googleapis.com/xtreme_translations/FLEURS102/${LANGUAGE_CODE}.tar.gz. - MasakhaNER is available from https://github.com/masakhane-io/masakhane-ner.
This is not an officially supported Google product.