Pretrained transformer-based models to predict toxicity in online comments, based on the Jigsaw Kaggle challenges. It includes a multilingual model trained on seven different languages.