Trained models & code to predict toxic comments on all 3 Jigsaw Toxic Comment Challenges. Built using ⚡ Pytorch Lightning and 🤗 Transformers.





GitHub Stars



Last Commit

21d ago









🙊 Detoxify

Toxic Comment Classification with ⚡ Pytorch Lightning and 🤗 Transformers

PyPI version GitHub all releases CI testing Lint

![Examples image](examples.png)

News & Updates

03-09-2021: New improved unbiased model

  • Updated the unbiased model weights used by Detoxify with a model trained on both datasets from the first 2 Jigsaw challenges. New best score on the test set: 0.93744 (0.93639 before).

14-01-2021: Lightweight models

  • Added smaller models trained with Albert for the original and unbiased models! Can access these in the same way with detoxify using original-small and unbiased-small as inputs. The original-small achieved a mean AUC score of 0.98281 (0.98636 before) and the unbiased-small achieved a final score of 0.93362 (0.93639 before).


Trained models & code to predict toxic comments on 3 Jigsaw challenges: Toxic comment classification, Unintended Bias in Toxic comments, Multilingual toxic comment classification.

Built by Laura Hanu at Unitary, where we are working to stop harmful content online by interpreting visual content in context.


  • For inference:
    • 🤗 Transformers
    • ⚡ Pytorch lightning
  • For training will also need:
    • Kaggle API (to download data)
ChallengeYearGoalOriginal Data SourceDetoxify Model NameTop Kaggle Leaderboard ScoreDetoxify Score
Toxic Comment Classification Challenge2018build a multi-headed model that’s capable of detecting different types of of toxicity like threats, obscenity, insults, and identity-based hate.Wikipedia Commentsoriginal0.988560.98636
Jigsaw Unintended Bias in Toxicity Classification2019build a model that recognizes toxicity and minimizes this type of unintended bias with respect to mentions of identities. You'll be using a dataset labeled for identity mentions and optimizing a metric designed to measure unintended bias.Civil Commentsunbiased0.947340.93744
Jigsaw Multilingual Toxic Comment Classification2020build effective multilingual modelsWikipedia Comments + Civil Commentsmultilingual0.95360.91655*

*Score not directly comparable since it is obtained on the validation set provided and not on the test set. To update when the test labels are made available.

It is also noteworthy to mention that the top leadearboard scores have been achieved using model ensembles. The purpose of this library was to build something user-friendly and straightforward to use.

Limitations and ethical considerations

If words that are associated with swearing, insults or profanity are present in a comment, it is likely that it will be classified as toxic, regardless of the tone or the intent of the author e.g. humorous/self-deprecating. This could present some biases towards already vulnerable minority groups.

The intended use of this library is for research purposes, fine-tuning on carefully constructed datasets that reflect real world demographics and/or to aid content moderators in flagging out harmful content quicker.

Some useful resources about the risk of different biases in toxicity or hate speech detection are:

Quick prediction

The multilingual model has been trained on 7 different languages so it should only be tested on: english, french, spanish, italian, portuguese, turkish or russian.

# install detoxify  

pip install detoxify

from detoxify import Detoxify

# each model takes in either a string or a list of strings

results = Detoxify('original').predict('example text')

results = Detoxify('unbiased').predict(['example text 1','example text 2'])

results = Detoxify('multilingual').predict(['example text','exemple de texte','texto de ejemplo','testo di esempio','texto de exemplo','örnek metin','пример текста'])

# to specify the device the model will be allocated on (defaults to cpu), accepts any torch.device input

model = Detoxify('original', device='cuda')

# optional to display results nicely (will need to pip install pandas)

import pandas as pd

print(pd.DataFrame(results, index=input_text).round(5))

For more details check the Prediction section.


All challenges have a toxicity label. The toxicity labels represent the aggregate ratings of up to 10 annotators according the following schema:

  • Very Toxic (a very hateful, aggressive, or disrespectful comment that is very likely to make you leave a discussion or give up on sharing your perspective)
  • Toxic (a rude, disrespectful, or unreasonable comment that is somewhat likely to make you leave a discussion or give up on sharing your perspective)
  • Hard to Say
  • Not Toxic

More information about the labelling schema can be found here.

Toxic Comment Classification Challenge

This challenge includes the following labels:

  • toxic
  • severe_toxic
  • obscene
  • threat
  • insult
  • identity_hate

Jigsaw Unintended Bias in Toxicity Classification

This challenge has 2 types of labels: the main toxicity labels and some additional identity labels that represent the identities mentioned in the comments.

Only identities with more than 500 examples in the test set (combined public and private) are included during training as additional labels and in the evaluation calculation.

  • toxicity
  • severe_toxicity
  • obscene
  • threat
  • insult
  • identity_attack
  • sexual_explicit

Identity labels used:

  • male
  • female
  • homosexual_gay_or_lesbian
  • christian
  • jewish
  • muslim
  • black
  • white
  • psychiatric_or_mental_illness

A complete list of all the identity labels available can be found here.

Jigsaw Multilingual Toxic Comment Classification

Since this challenge combines the data from the previous 2 challenges, it includes all labels from above, however the final evaluation is only on:

  • toxicity

How to run

First, install dependencies

# clone project   

git clone

# create virtual env

python3 -m venv toxic-env
source toxic-env/bin/activate

# install project   

pip install -e detoxify
cd detoxify

# for training
pip install -r requirements.txt


Trained models summary:

Model nameTransformer typeData from
originalbert-base-uncasedToxic Comment Classification Challenge
unbiasedroberta-baseUnintended Bias in Toxicity Classification
multilingualxlm-roberta-baseMultilingual Toxic Comment Classification

For a quick prediction can run the example script on a comment directly or from a txt containing a list of comments.

# load model via torch.hub

python --input 'example' --model_name original

# load model from from checkpoint path

python --input 'example' --from_ckpt_path model_path

# save results to a .csv file

python --input test_set.txt --model_name original --save_to results.csv

# to see usage

python --help

Checkpoints can be downloaded from the latest release or via the Pytorch hub API with the following names:

  • toxic_bert
  • unbiased_toxic_roberta
  • multilingual_toxic_xlm_r
model = torch.hub.load('unitaryai/detoxify','toxic_bert')

Importing detoxify in python:

from detoxify import Detoxify

results = Detoxify('original').predict('some text')

results = Detoxify('unbiased').predict(['example text 1','example text 2'])

results = Detoxify('multilingual').predict(['example text','exemple de texte','texto de ejemplo','testo di esempio','texto de exemplo','örnek metin','пример текста'])

# to display results nicely

import pandas as pd



If you do not already have a Kaggle account:

  • you need to create one to be able to download the data

  • go to My Account and click on Create New API Token - this will download a kaggle.json file

  • make sure this file is located in ~/.kaggle

create data directory

mkdir jigsaw_data cd jigsaw_data

download data

kaggle competitions download -c jigsaw-toxic-comment-classification-challenge

kaggle competitions download -c jigsaw-unintended-bias-in-toxicity-classification

kaggle competitions download -c jigsaw-multilingual-toxic-comment-classification

## Start Training
 ### Toxic Comment Classification Challenge



python --config configs/Toxic_comment_classification_BERT.json

Unintended Bias in Toxicicity Challenge

python --config configs/Unintended_bias_toxic_comment_classification_RoBERTa.json

Multilingual Toxic Comment Classification

This is trained in 2 stages. First, train on all available data, and second, train only on the translated versions of the first challenge.

The translated data can be downloaded from Kaggle in french, spanish, italian, portuguese, turkish, and russian (the languages available in the test set).

# stage 1

python --config configs/Multilingual_toxic_comment_classification_XLMR.json

# stage 2

python --config configs/Multilingual_toxic_comment_classification_XLMR_stage2.json --resume path_to_saved_checkpoint_stage1

Monitor progress with tensorboard

tensorboard --logdir=./saved

Model Evaluation

Toxic Comment Classification Challenge

This challenge is evaluated on the mean AUC score of all the labels.

python --checkpoint saved/lightning_logs/checkpoints/example_checkpoint.pth --test_csv test.csv

Unintended Bias in Toxicicity Challenge

This challenge is evaluated on a novel bias metric that combines different AUC scores to balance overall performance. More information on this metric here.

python --checkpoint saved/lightning_logs/checkpoints/example_checkpoint.pth --test_csv test.csv

# to get the final bias metric
python model_eval/

Multilingual Toxic Comment Classification

This challenge is evaluated on the AUC score of the main toxic label.

python --checkpoint saved/lightning_logs/checkpoints/example_checkpoint.pth --test_csv test.csv


  author={Hanu, Laura and {Unitary team}},

Rate & Review

Great Documentation0
Easy to Use0
Highly Customizable0
Bleeding Edge0
Responsive Maintainers0
Poor Documentation0
Hard to Use0
Unwelcoming Community0