app.py

import gradio as gr
import spaces
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
import torch
import logging

# Configure logging/logger
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger(__name__)

# Stores for models and tokenizers
tokenizers = {}
pipelines = {}

# Predefined list of models to compare (can be expanded)
model_options = {
    "Foundation-Sec-8B": "fdtn-ai/Foundation-Sec-8B",
}

# Initialize models at startup
for model_name, model_path in model_options.items():
    try:
        logger.info(f"Initializing text generation model: {model_path}")
        tokenizers[model_path] = AutoTokenizer.from_pretrained(model_path)
        pipelines[model_path] = pipeline(
            "text-generation",
            model=model_path,
            tokenizer=tokenizers[model_path],
            torch_dtype=torch.bfloat16,
            device_map="auto",
            trust_remote_code=True
        )
        logger.info(f"Model initialized successfully: {model_path}")
    except Exception as e:
        logger.error(f"Error initializing model {model_path}: {str(e)}")

@spaces.GPU
def generate_text_local(model_path, prompt, max_new_tokens=512, temperature=0.7, top_p=0.95):
    """Local text generation"""
    try:
        # Use the already initialized model
        if model_path in pipelines:
            model_pipeline = pipelines[model_path]
            logger.info(f"Running text generation with {model_path}")

            outputs = model_pipeline(
                prompt,
                max_new_tokens=max_new_tokens,
                do_sample=True,
                temperature=temperature,
                top_p=top_p,
                clean_up_tokenization_spaces=True,
            )

            return outputs[0]["generated_text"].replace(prompt, "").strip()
        else:
            return f"Error: Model {model_path} not initialized"
    except Exception as e:
        logger.error(f"Error in text generation with {model_path}: {str(e)}")
        return f"Error: {str(e)}"

# Build Gradio app
def create_demo():
    with gr.Blocks() as demo:
        gr.Markdown("# AI Model Comparison Tool 🌟")
        gr.Markdown(
            """
            Compare responses from two AI models side-by-side.  
            Select two models, ask a question, and compare their responses in real time!
            """
        )

        # Input Section
        with gr.Row():
            system_message = gr.Textbox(
                value="You are a helpful assistant providing answers for technical and customer support queries.",
                label="System message"
            )
            user_message = gr.Textbox(label="Your question", placeholder="Type your question here...")

        with gr.Row():
            max_tokens = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens")
            temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
            top_p = gr.Slider(
                minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"
            )

        # Model Selection Section
        selected_models = gr.CheckboxGroup(
            choices=list(model_options.keys()),
            label="Select exactly two model to compare",
            value=["Foundation-Sec-8B"],  # Default models
        )

        # Dynamic Response Section
        response_box1 = gr.Textbox(label="Response from Model 1", interactive=False)
        #response_box2 = gr.Textbox(label="Response from Model 2", interactive=False)

        # Function to generate responses
        def generate_responses(
            message, system_message, max_tokens, temperature, top_p, selected_models
        ):
            #if len(selected_models) != 2:
            #    return "Error: Please select exactly two models to compare.", ""
            
            if len(selected_models) == 0:
                return "Error: Please select at least one model"
            
            model_path = model_options[selected_models[0]]
            full_prompt = f"{system_message}\n\nUser: {message}\nAssistant:"
            response = generate_text_local(
                model_path,
                full_prompt,
                max_tokens,
                temperature,
                top_p
            )
            
            #return responses.get(selected_models[0], ""), responses.get(selected_models[1], "")
            return response
            
        # Add a button for generating responses
        submit_button = gr.Button("Generate Responses")
        submit_button.click(
            generate_responses,
            inputs=[user_message, system_message, max_tokens, temperature, top_p, selected_models],
            #outputs=[response_box1, response_box2],  # Link to response boxes
            outputs=[response_box1]
        )

    return demo

if __name__ == "__main__":
    demo = create_demo()
    demo.launch()