# unitaryai/detoxify

Trained models & code to predict toxic comments on all 3 Jigsaw Toxic Comment Challenges. Built using ⚡ Pytorch Lightning and 🤗 Transformers. For access to our API, please email us at contact@unitary.ai.

Repository: https://github.com/unitaryai/detoxify
Canonical: https://ross.abutalabs.com/products/detoxify
Homepage: https://www.unitary.ai/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: bert, bert-model, huggingface-transformers, huggingface, nlp, toxic-comment-classification, toxicity, toxic-comments, sentence-classification, kaggle-competition, pytorch-lightning, hatespeech, hate-speech-detection, toxicity-classification, hate-speech
Last push: 2026-07-06T21:15:21+00:00

## Health v2 (maintenance only)
Score: 76/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 91, release rhythm 44, longevity 100
- inputs: {"age_days": 2170, "days_push": 58, "days_rel": 160, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1291, forks 143 (observed 2026-08-28T04:04:15.661675+00:00)

## What it is
Detoxify is a Python library providing pre-trained transformer models that classify toxic comments, built on PyTorch Lightning and Hugging Face Transformers. It offers models trained on the three Jigsaw Kaggle toxic comment challenges, including original, unbiased, and multilingual variants.

## Use cases
- detect toxic comments in user-generated text
- moderate hate speech on a forum
- classify insults, threats, and identity attacks in comments
- screen multilingual comments for toxicity
- add a toxicity score to a chat moderation pipeline
- benchmark against Jigsaw Kaggle competition models

## When to choose
- you need ready-made pre-trained toxicity classifiers in Python
- you want models with strong Jigsaw competition leaderboard performance
- you need multilingual or bias-reduced toxicity detection
- you want a lightweight pip-installable inference library

## When to avoid
- you need real-time content moderation at massive scale with an API service
- you need toxicity detection for images or video rather than text
- you need custom toxicity categories not covered by the Jigsaw labels
- you require frequent updates or active feature development

## Facets
- artifact type: library
- maturity: maintenance
- function: nlp, machine-learning, deep-learning
- domain: machine-learning, artificial-intelligence
- platform: python
- tags: toxicity-detection, hate-speech-detection, content-moderation, transformers, pytorch-lightning, bert, text-classification, jigsaw, natural-language-processing

## Member repositories
- unitaryai/detoxify (main) score 76

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:15.661675+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:54:49.692207+00:00, confidence not recorded.
  - readme: https://github.com/unitaryai/detoxify (fetched 2026-08-28T04:04:15.661675+00:00, sha d58691b30cb4)
  - homepage: https://www.unitary.ai/ (fetched 2026-08-29T12:11:23.987330+00:00, sha b3b5557cd6ea)
  - site_page: https://www.unitary.ai/about-us (fetched 2026-08-29T12:11:23.999662+00:00, sha 964612ffa876)
  - registry_pypi: https://pypi.org/pypi/detoxify/json (fetched 2026-08-29T12:11:24.001797+00:00, sha 8e587004dadf)
  - site_page: https://www.unitary.ai/pricing (fetched 2026-08-29T12:11:23.996544+00:00, sha 797886befae4)
- Data as of 2026-08-30T08:39:29.467469+00:00.