app.py

import gradio as gr
import os
from evaluation_logic import run_evaluation
from eval.predict import PROMPT_FORMATTERS

PROMPT_TEMPLATES = {
    "duckdbinstgraniteshort": PROMPT_FORMATTERS["duckdbinstgraniteshort"]().PROMPT_TEMPLATE,
    "duckdbinst": PROMPT_FORMATTERS["duckdbinst"]().PROMPT_TEMPLATE,
}

def gradio_run_evaluation(inference_api, model_name, prompt_format, openrouter_token=None, custom_prompt=None):
    # Set environment variable if OpenRouter token is provided
    if inference_api == "openrouter":
        os.environ["OPENROUTER_API_KEY"] = str(openrouter_token)

    output = []
    for result in run_evaluation(inference_api, str(model_name).strip(), prompt_format, custom_prompt):
        output.append(result)
        yield "\n".join(output)

def update_token_visibility(api):
    """Update visibility of the OpenRouter token input"""
    return gr.update(visible=api == "openrouter")

def on_prompt_format_change(choice):
    """Only update content when selecting a preset"""
    if choice in PROMPT_TEMPLATES:
        return PROMPT_TEMPLATES[choice]
    return gr.update()

with gr.Blocks(gr.themes.Soft()) as demo:
    gr.Markdown("# DuckDB SQL Evaluation App")

    with gr.Row():
        with gr.Column():
            inference_api = gr.Dropdown(
                label="Inference API",
                choices=['openrouter'],
                value="openrouter"
            )

            openrouter_token = gr.Textbox(
                label="OpenRouter API Token",
                placeholder="Enter your OpenRouter API token",
                type="password",
                visible=True
            )

            model_name = gr.Textbox(
                label="Model Name (e.g., qwen/qwen-2.5-72b-instruct)"
            )

            gr.Markdown("[View OpenRouter Models](https://openrouter.ai/models?order=top-weekly)")

    with gr.Row():
        with gr.Column():
            prompt_format = gr.Dropdown(
                label="Prompt Format",
                choices=['duckdbinst', 'duckdbinstgraniteshort', 'custom'],
                value="duckdbinstgraniteshort"
            )

            custom_prompt = gr.TextArea(
                label="Prompt Template Content",
                placeholder="Enter your custom prompt template here or select a preset format above.",
                lines=10,
                value=PROMPT_TEMPLATES['duckdbinstgraniteshort']
            )

    gr.Examples(
        examples=[
            ["openrouter", "qwen/qwen-2.5-72b-instruct", "duckdbinst"],
            ["openrouter", "meta-llama/llama-3.2-3b-instruct:free", "duckdbinstgraniteshort"],
            ["openrouter", "mistralai/mistral-nemo", "duckdbinst"],
        ],
        inputs=[inference_api, model_name, prompt_format],
    )

    start_btn = gr.Button("Start Evaluation")
    output = gr.Textbox(label="Output", lines=20)

    # Event handlers
    inference_api.change(
        fn=update_token_visibility,
        inputs=[inference_api],
        outputs=[openrouter_token]
    )

    prompt_format.change(
        fn=on_prompt_format_change,
        inputs=[prompt_format],
        outputs=[custom_prompt]
    )

    custom_prompt.input(
        fn=lambda: "custom",
        outputs=prompt_format
    )

    start_btn.click(
        fn=gradio_run_evaluation,
        inputs=[inference_api, model_name, prompt_format, openrouter_token, custom_prompt],
        outputs=output
    )

demo.queue().launch()