toxicity-classifier-2

Toxicity Classifier

A fine-tuned DistilRoBERTa model for detecting toxic comments in text. Trained on Civil Comments and Real Toxicity Prompts datasets with balanced class distribution (50k toxic + 50k non-toxic examples).

Model Details

• Base Model: distilroberta-base
• Task: Binary text classification (Toxic / Non-toxic)
• Training Data: 100k balanced samples from Civil Comments + Real Toxicity Prompts
• Accuracy: 90.88%
• Precision: 90.50%
• Recall: 91.16%
• F1-Score: 90.83%

To Load:

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model = AutoModelForSequenceClassification.from_pretrained("Jaanhavi/toxicity-classifier")
tokenizer = AutoTokenizer.from_pretrained("Jaanhavi/toxicity-classifier")

To Use (Simple Pipeline):

from transformers import pipeline

classifier = pipeline("text-classification", model="Jaanhavi/toxicity-classifier")

text = "You are so stupid and worthless"
result = classifier(text)
print(result)
# Output: [{'label': 'LABEL_1', 'score': 0.9980}]
# LABEL_0 = Non-toxic, LABEL_1 = Toxic

Label Meanings

• LABEL_0: Non-toxic text
• LABEL_1: Toxic text (includes insults, threats, harmful content)

Performance

Tested on 10,000 diverse examples with 90.88% overall accuracy.

Examples:

• "you are so stupid" → TOXIC ✅
• "I love your work" → NON-TOXIC ✅
• "go commit suicide" → TOXIC ✅
• "have a nice day" → NON-TOXIC ✅