app.py

import torch
import gradio as gr
import numpy as np
import matplotlib.pyplot as plt
from transformers import AutoTokenizer, AutoModelForSequenceClassification

torch.set_num_threads(torch.get_num_threads())

# Load the trained model and tokenizer from Hugging Face Hub
model_path = "HyperX-Sentience/RogueBERT-Toxicity-85K"
model = AutoModelForSequenceClassification.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained(model_path)

# Move the model to CUDA if available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Define toxicity labels
labels = ["toxic", "severe_toxic", "obscene", "threat", "insult", "identity_hate"]

def predict_toxicity(comment):
    """Predicts the toxicity levels of a given comment."""
    inputs = tokenizer(comment, truncation=True, padding="max_length", max_length=128, return_tensors="pt")
    inputs = {key: val.to(device) for key, val in inputs.items()}
    
    with torch.no_grad():
        outputs = model(**inputs)
        probabilities = torch.sigmoid(outputs.logits).cpu().numpy()[0]
    
    return {labels[i]: float(probabilities[i]) for i in range(len(labels))}

def visualize_toxicity(comment):
    """Generates a bar chart showing toxicity levels."""
    scores = predict_toxicity(comment)
    
    # Create bar chart
    plt.figure(figsize=(6, 4))
    plt.bar(scores.keys(), scores.values(), color=['blue', 'red', 'green', 'purple', 'orange', 'brown'])
    plt.ylim(0, 1)
    plt.ylabel("Toxicity Score")
    plt.title("Toxicity Analysis")
    plt.xticks(rotation=45)
    plt.grid(axis='y', linestyle='--', alpha=0.7)
    
    # Save plot to display in Gradio
    plt.savefig("toxicity_plot.png")
    plt.close()
    
    return "toxicity_plot.png"

# Gradio interface
demo = gr.Interface(
    fn=visualize_toxicity,
    inputs=gr.Textbox(label="Enter a comment:"),
    outputs=gr.Image(type="file", label="Toxicity Scores"),
    title="Toxicity Detection with RogueBERT",
    description="Enter a comment to analyze its toxicity levels. The results will be displayed as a bar chart."
)

demo.launch()