app.py

import os
import gradio as gr
from huggingface_hub import login, InferenceClient
import spaces

# Authenticate with Hugging Face API
api_key = os.getenv("TOKEN")
login(api_key)

# Predefined list of models to compare (can be expanded)
model_options = {
    "Llama-3.1-70B": "meta-llama/Llama-3.1-70B-Instruct",
    "Qwen-2.5-1.5B-Instruct": "Qwen/Qwen2.5-1.5B-Instruct",
    "Llama-3.2-1B": "meta-llama/Llama-3.2-1B",
    "DeepSeek-V2.5": "deepseek-ai/DeepSeek-V2.5",
    "Athene-V2-Chat": "Nexusflow/Athene-V2-Chat",
}

# Initialize clients for models
clients = {name: InferenceClient(repo_id) for name, repo_id in model_options.items()}

# Define the response function
@spaces.GPU
def respond(
    message,
    history: list[dict],
    system_message,
    max_tokens,
    temperature,
    top_p,
    selected_models,
):
    messages = [{"role": "system", "content": system_message}] + history
    messages.append({"role": "user", "content": message})

    responses = {}

    # Generate responses for each selected model
    for model_name in selected_models:
        client = clients[model_name]
        response = ""
        for token in client.chat_completion(
            messages, max_tokens=max_tokens, stream=True, temperature=temperature, top_p=top_p
        ):
            delta = token.choices[0].delta.content
            response += delta
        responses[model_name] = response

    return responses

# 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 models to compare",
            value=["Llama-3.1-70B", "Qwen-2.5-1.5B-Instruct"],  # 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.", ""
            responses = respond(
                message, [], system_message, max_tokens, temperature, top_p, selected_models
            )
            return responses.get(selected_models[0], ""), responses.get(selected_models[1], "")

        # 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
        )

    return demo

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