app.py

# 重构后的 app.py
import gradio as gr
from utils.grace_plot import plot_radar
from model_wrappers.model_a import run_model_a
from model_wrappers.model_b import run_model_b

def run_all_models(prompt):
    return run_model_a(prompt), run_model_b(prompt)

with gr.Blocks() as demo:
    with gr.Tab("LLM Benchmark"):
        gr.Markdown("## ✨ 模型 GRACE 维度雷达图")
        with gr.Row():
            plot_btn = gr.Button("生成 GRACE 雷达图")
        radar_output = gr.Plot()
        plot_btn.click(fn=plot_radar, inputs=[], outputs=radar_output)

    with gr.Tab("Arena"):
        gr.Markdown("## 🤖 模型竞技场：同一输入比拼")
        prompt = gr.Textbox(label="请输入 Prompt")
        run_btn = gr.Button("运行所有模型")
        output_a = gr.Textbox(label="Model A 输出")
        output_b = gr.Textbox(label="Model B 输出")
        run_btn.click(fn=run_all_models, inputs=[prompt], outputs=[output_a, output_b])

    with gr.Tab("Report"):
        with open("report.md", encoding="utf-8") as f:
            report_md = f.read()
        gr.Markdown(report_md)

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