app.py

import gradio as gr



from src.dataloading import get_leaderboard_models_cached, get_leaderboard_datasets


def create_demo():
    # Fetch data once on startup (cache this in production)
    models = get_leaderboard_models_cached()
    datasets = get_leaderboard_datasets()

    with gr.Blocks(title="LLM Similarity Analyzer") as demo:
        gr.Markdown("## Compare Models/Datasets from Open LLM Leaderboard")
        
        with gr.Row():
            model_dropdown = gr.Dropdown(
                choices=models,
                label="Select Model",
                filterable=True,
                interactive=True,
                allow_custom_value=False,
                info="Search models from Open LLM Leaderboard"
            )
            
            dataset_dropdown = gr.Dropdown(
                choices=datasets,
                label="Select Dataset",
                filterable=True,
                interactive=True,
                info="Leaderboard benchmark datasets"
            )

        # Add your similarity computation and visualization components here
        # Example placeholder:
        similarity_output = gr.Textbox(label="Similarity Score")
        compute_btn = gr.Button("Compute Similarity")
        
        def compute_similarity(model, dataset):
            # Replace with your actual similarity metric
            return f"Similarity between {model} and {dataset}: {0.85:.2f}"
        
        compute_btn.click(
            fn=compute_similarity,
            inputs=[model_dropdown, dataset_dropdown],
            outputs=similarity_output
        )
        
    return demo


def create_demo_with_refresh():
    demo = create_demo()
    
    with demo:
        refresh_btn = gr.Button("Refresh Model List")
        def refresh_models():
            return gr.Dropdown(choices=get_leaderboard_models_cached())
        
        refresh_btn.click(
            fn=refresh_models,
            outputs=model_dropdown
        )
    
    return demo

demo = create_demo_with_refresh()
demo.launch()