Skip to content

Dataset · google

XTREME multilingual benchmark

google/xtreme

Multilingual pairs for NLI and other tasks: 112.5k annotated pairs across 14 languages.

What's inside

The card describes XNLI: 5,000 test and 2,500 dev pairs from MultiNLI with textual entailment labeling, translated into 14 languages including Russian, Arabic, Chinese, and Hindi. A total of 112.5k annotated pairs are specified. The metadata also lists multiple-choice, question-answering, token-classification, text-classification, and text-retrieval tasks; the format is not specified in the card.

What you can do with it

  • Evaluate cross-lingual NLI
  • Fine-tune a multilingual text classifier
  • Check zero-shot transfer to other languages
Tasks
Multiple choice, Question answering, Token classification, Text classification, Text retrieval
Languages
Afrikaans, Arabic, Bulgarian, Bangla, German, Greek, English, Spanish and 4 more
Size
1M–10M rows
Downloads this month
29.1K
Likes
119

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("google/xtreme")