app_hf.py

import gradio as gr
from inference import ContentClassifierInference
import os
import json
import logging

# Setup logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Initialize the model
try:
    model = ContentClassifierInference()
    model_initialized = True
    logger.info("Model initialized successfully")
except Exception as e:
    logger.error(f"Error initializing model: {e}")
    model_initialized = False

def classify_text(text):
    """Classify single text input"""
    if not model_initialized:
        return json.dumps({"error": "Model initialization failed"}, indent=2)
    
    if not text or not text.strip():
        return json.dumps({"error": "Please provide valid text input"}, indent=2)
    
    try:
        result = model.predict(text.strip())
        logger.info(f"Processed text classification: {result['threat_prediction']}")
        return json.dumps(result, indent=2)
    except Exception as e:
        logger.error(f"Classification error: {e}")
        return json.dumps({"error": str(e)}, indent=2)

def classify_batch(text):
    """Classify batch of texts (one per line)"""
    if not model_initialized:
        return json.dumps({"error": "Model initialization failed"}, indent=2)
    
    if not text or not text.strip():
        return json.dumps({"error": "Please provide valid text input"}, indent=2)
    
    try:
        # Split by newlines for batch processing
        texts = [t.strip() for t in text.split("\n") if t.strip()]
        if not texts:
            return json.dumps({"error": "No valid texts provided"}, indent=2)
            
        if len(texts) > 10:  # Limit batch size
            return json.dumps({"error": "Batch size limited to 10 texts"}, indent=2)
            
        results = model.predict_batch(texts)
        logger.info(f"Processed batch of {len(texts)} texts")
        return json.dumps(results, indent=2)
    except Exception as e:
        logger.error(f"Batch classification error: {e}")
        return json.dumps({"error": str(e)}, indent=2)

# API function for external use
def predict_api(text):
    """API endpoint for programmatic access"""
    if isinstance(text, list):
        return [json.loads(classify_text(t)) for t in text]
    else:
        return json.loads(classify_text(text))

# Create Gradio interface
with gr.Blocks(title="Content Classifier", theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🔍 Content Classifier")
    gr.Markdown("""
    This tool classifies content as either **safe** or **unsafe** using an ONNX model.  
    Perfect for content moderation, safety checks, and automated text analysis.
    """)
    
    # Status indicator
    if model_initialized:
        gr.Markdown("✅ **Model Status**: Ready")
    else:
        gr.Markdown("❌ **Model Status**: Failed to initialize")
    
    with gr.Tab("Single Text Classification"):
        with gr.Row():
            with gr.Column():
                text_input = gr.Textbox(
                    label="Enter text to classify", 
                    lines=5, 
                    placeholder="Type or paste your text here...",
                    max_lines=10
                )
                classify_btn = gr.Button("🔍 Classify", variant="primary", size="lg")
                
                # Examples
                gr.Examples(
                    examples=[
                        ["This is a normal, safe piece of content."],
                        ["Hello, how are you doing today?"],
                        ["Example text for content classification"]
                    ],
                    inputs=text_input,
                    label="Try these examples:"
                )
                
            with gr.Column():
                result_output = gr.JSON(label="Classification Result", show_label=True)
        classify_btn.click(fn=classify_text, inputs=text_input, outputs=result_output)
    
    with gr.Tab("Batch Processing"):
        with gr.Row():
            with gr.Column():
                batch_input = gr.Textbox(
                    label="Enter multiple texts (one per line)", 
                    lines=10, 
                    placeholder="Text 1\nText 2\nText 3\n...(max 10 texts)",
                    max_lines=15
                )
                batch_btn = gr.Button("📋 Process Batch", variant="primary", size="lg")
                
                gr.Markdown("**Note**: Maximum 10 texts per batch")
                
            with gr.Column():
                batch_output = gr.JSON(label="Batch Classification Results", show_label=True)
        
        batch_btn.click(fn=classify_batch, inputs=batch_input, outputs=batch_output)
    
    with gr.Tab("API Documentation"):
        gr.Markdown("""
        ## 🔌 API Usage
        
        This Space can be used as an API endpoint for programmatic access.
        
        ### Single Text Classification
        ```python
        import requests
        
        url = "https://your-space-name.hf.space/predict"
        response = requests.post(url, json={"text": "Your content to classify"})
        result = response.json()
        ```
        
        ### Batch Processing
        ```python
        import requests
        
        url = "https://your-space-name.hf.space/predict"
        texts = ["Text 1", "Text 2", "Text 3"]
        response = requests.post(url, json={"text": texts})
        results = response.json()
        ```
        
        ### Response Format
        ```json
        {
          "is_threat": false,
          "final_confidence": 0.85,
          "threat_prediction": "safe",
          "onnx_prediction": {
            "safe": 0.85,
            "unsafe": 0.15
          },
          "models_used": ["onnx"],
          "raw_predictions": {...}
        }
        ```
        
        ### Using with curl
        ```bash
        curl -X POST https://your-space-name.hf.space/predict \\
          -H "Content-Type: application/json" \\
          -d '{"text": "Your content to classify"}'
        ```
        """)

# Launch the app
if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860, share=False)