marabert2-levantine-toxic-model-v4

Levantine Hate Speech Classifier

This model identifies Hate Speech and Abusive Language in Levantine Arabic tweets. It was fine-tuned on the dataset: alan9622/marabert2-levantine-toxic-model-v4-dataset.

📊 Evaluation

Metric	Score
Accuracy	0.9128205128205128
F1 Score	0.908218073979278

📖 Labels

ID	Label	Definition
0	normal	Safe text.
1	toxic	abusive or hate Insults/Vulgarity not directed at identity or Attacks based on religion, sect, gender, or nationality.

Usage

from transformers import pipeline
classifier = pipeline("text-classification", model="alan9622/marabert2-levantine-toxic-model-v4")
classifier("يا زلمة حل عني")

Downloads last month: 9

Safetensors

Model size

0.2B params

Tensor type

F32

Model tree for alan9622/marabert2-levantine-toxic-model-v4

Base model

UBC-NLP/MARBERTv2

Finetuned

(39)

this model

Dataset used to train alan9622/marabert2-levantine-toxic-model-v4

Evaluation results

accuracy on L-HSAB Custom
self-reported

0.913
f1 on L-HSAB Custom
self-reported

0.908