Update app.py

app.py

@@ -1,101 +1,45 @@
 import gradio as gr
-import torch
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
-import easyocr
-from PIL import Image
-import numpy as np
-
-# -------------------------------
-# MODEL: CardiffNLP RoBERTa Hate Classifier
-# -------------------------------
-MODEL_NAME = "cardiffnlp/twitter-roberta-base-hate-multiclass-latest"
-tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
-
-LABELS = ["sexism", "racism", "disability", "sexual_orientation", "religion", "other", "not_hate"]
+import torch
 
-# -------------------------------
-# OCR Setup
-# -------------------------------
-reader = easyocr.Reader(["en"], gpu=False)
+# Using Microsoft DeBERTa v3 base model (general-purpose, fine-tune recommended)
+MODEL_NAME = "microsoft/deberta-v3-base"
+LABELS = [
+    "sexism",
+    "racism",
+    "disability",
+    "sexual_orientation",
+    "religion",
+    "other",
+    "not_hate"
+]
 
-def extract_text(image):
-    """Extract text from uploaded image using EasyOCR"""
-    if image is None:
-        return ""
-    if isinstance(image, Image.Image):
-        image = np.array(image)
-    result = reader.readtext(image, detail=0)
-    return " ".join(result)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME, num_labels=len(LABELS))
 
-# -------------------------------
-# CLASSIFICATION LOGIC
-# -------------------------------
 def classify_text(text):
-    """Classify input text using RoBERTa model"""
-    if not text.strip():
-        return "No text found for analysis.", None
-    inputs = tokenizer(text, return_tensors="pt", truncation=True)
+    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
     with torch.no_grad():
-        logits = model(**inputs).logits
+        outputs = model(**inputs)
+        logits = outputs.logits
         probs = torch.nn.functional.softmax(logits, dim=-1)
         pred = torch.argmax(probs).item()
         confidence = float(probs[0][pred])
-        label = LABELS[pred]
-    return f"Category: {label} (Confidence: {confidence:.2f})", label
-
-
-# -------------------------------
-# CHATBOT FUNCTION
-# -------------------------------
-def cyberbully_chat(messages, user_message, image=None):
-    """Chat-like conversational function"""
-    history = messages or []
-
-    # Extract text from image (if any)
-    if image:
-        text = extract_text(image)
-        content = f"[Extracted from image] {text}" if text else "[No readable text found]"
-    else:
-        text = user_message
-        content = text.strip()
-
-    if not content or content == "[No readable text found]":
-        history.append({"role": "assistant", "content": "Please provide valid text or an image with text."})
-        return history
-
-    # Classify with model
-    classification, label = classify_text(text)
-    
-    # Append to chat
-    history.append({"role": "user", "content": content})
-    history.append({"role": "assistant", "content": classification})
-    return history
-
-
-# -------------------------------
-# GRADIO ChatGPT-like UI
-# -------------------------------
-with gr.Blocks() as demo:
-    gr.Markdown("# 🤖 Cyber Bully Detection System")
-    gr.Markdown("Upload an image or type text. The system will analyze hate-speech categories using a RoBERTa model specialized for social media context.")
-
-    chatbot = gr.Chatbot(type="messages", label="CyberBully Chat")
-    with gr.Row():
-        text_input = gr.Textbox(show_label=False, placeholder="Type a message here...")
-        image_input = gr.Image(source="upload", type="pil", label="Upload Screenshot (optional)")
-    with gr.Row():
-        submit_btn = gr.Button("Analyze")
-        clear_btn = gr.Button("Clear Chat")
-
-    submit_btn.click(
-        cyberbully_chat,
-        [chatbot, text_input, image_input],
-        [chatbot],
-        queue=True
-    )
-
-    clear_btn.click(lambda: [], None, chatbot, queue=False)
+        return LABELS[pred], confidence
+
+def chatbot(text):
+    if not text or not text.strip():
+        return "Please enter some text."
+    label, confidence = classify_text(text)
+    return f"Prediction: {label} (Confidence: {confidence:.2f})"
+
+iface = gr.Interface(
+    fn=chatbot,
+    inputs=gr.Textbox(lines=3, placeholder="Enter text for hate speech classification"),
+    outputs="text",
+    title="DeBERTa Hate Speech Classifier",
+    description="Classifies text into hate speech categories with DeBERTa v3-base model."
+)
 
 if __name__ == "__main__":
-    demo.launch()
+    iface.launch()