Skip to content

Dataset · textdetox

Multilingual toxicity detection dataset (including Russian)

textdetox/multilingual_toxicity_dataset

A balanced corpus of texts in over ten languages for training and evaluating toxicity classifiers.

What's inside

This dataset is designed for binary toxicity classification and was prepared for the TextDetox 2024 and 2025 competitions. It contains 10K to 100K texts in Russian, English, Ukrainian, German, Spanish, Italian, French, Japanese, Tatar, and other languages. Each language in the 2024 version includes 5,000 balanced examples, split evenly between toxic and non-toxic.

What you can do with it

  • Training multilingual and Russian content moderation classifiers
  • Building safety filters for user prompts and LLM responses
  • Evaluating toxicity detection models on a balanced benchmark
Tasks
Text classification
Languages
English, Russian, Ukrainian, German, Spanish, Amharic, Chinese, Arabic and 4 more
Size
10K–100K rows
Downloads this month
1.4K
Likes
38

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("textdetox/multilingual_toxicity_dataset")