BatsResearch/cross-lingual-detox

Code for "Preference Tuning For Toxicity Mitigation Generalizes Across Languages." Paper accepted at Findings of EMNLP 2024

22
/ 100
Experimental

This project helps large language model (LLM) developers make their models generate less toxic content, even across different languages. By feeding the tool examples of preferred, less toxic English text, it fine-tunes the LLM to reduce harmful outputs. The output is a modified LLM that produces safer content in various languages without needing specific training for each one, benefiting developers building multilingual AI applications.

No commits in the last 6 months.

Use this if you are an LLM developer aiming to reduce toxicity in your models' outputs across multiple languages with a single English-based fine-tuning process.

Not ideal if you are an end-user of an LLM and want to filter out toxic content from an existing model without direct access to its development or fine-tuning process.

LLM development content moderation multilingual AI AI safety model fine-tuning
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 6 / 25
Maturity 16 / 25
Community 0 / 25

How are scores calculated?

Stars

18

Forks

Language

Jupyter Notebook

License

BSD-3-Clause

Last pushed

Mar 25, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/nlp/BatsResearch/cross-lingual-detox"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.