citizenlab/distilbert-base-multilingual-cased-toxicity

This is multilingual Distil-Bert model sequence classifier trained based on JIGSAW Toxic Comment Classification Challenge dataset.

How to use it

from transformers import pipeline

model_path = "citizenlab/distilbert-base-multilingual-cased-toxicity"

toxicity_classifier = pipeline("text-classification", model=model_path, tokenizer=model_path)
toxicity_classifier("this is a lovely message")
> [{'label': 'not_toxic', 'score': 0.9954179525375366}]

toxicity_classifier("you are an idiot and you and your family should go back to your country")
> [{'label': 'toxic', 'score': 0.9948776960372925}]

Evaluation

Accuracy

  Accuracy Score = 0.9425
F1 Score (Micro) = 0.9450549450549449
F1 Score (Macro) = 0.8491432341169309