# -*- coding: utf-8 -*-
import gradio as gr


from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

# Load a general Arabic sentiment analysis model as a proxy for offensive language detection
model_name = "nassga/arabic-offensive-comment-model"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Create pipeline for classification
classifier = pipeline("text-classification", model=model, tokenizer=tokenizer)

def predict_offensive(text):
    result = classifier(text)[0]
    label = "Offensive" if result["label"] == "LABEL_1" else "Not Offensive"
    return f"{label} (Confidence: {result['score']:.2f})"

gr.Interface(
    fn=predict_offensive,
    inputs=gr.Textbox(label="Enter a Kuwaiti Arabic Comment"),
    outputs=gr.Text(label="Prediction"),
    title="Kuwaiti Arabic Offensive Language Detector",
    description="Detects whether a comment is offensive using a fine-tuned Arabic BERT model."
).launch()