import gradio as gr
import pandas as pd

file1 = '综合排名.csv'
df1 = pd.read_csv(file1)
file2 ='basic-leaderboard.csv'
df2 = pd.read_csv(file2)
file3 ='学科.csv'
df3 = pd.read_csv(file3)
file4 ='安全.csv'
df4 = pd.read_csv(file4)
file5 ='大学试题.csv'
df5 = pd.read_csv(file5)
file6 ='中学试题.csv'
df6 = pd.read_csv(file6)
file7 ='通用语言能力排行榜-大模型裁判.csv'
df7 = pd.read_csv(file7)

def show_general_leaderboard():
    return df1

def show_basic_leaderboard():
    return df2

def show_subject_leaderboard():
    return df3

def show_safety_leaderboard():
    return df4

def show_college_leaderboard():
    return df5

def show_midschool_leaderboard():
    return df6

def show_LLM_judge_leaderboard():
    return df7


with gr.Blocks() as demo:
    gr.Markdown(
            """
            # Large Language Model Assessment in the Chinese Context / 中文语境下的人工智能大语言模型评测  
            by Zhenhui(Jack) Jiang, Jiaxin Li, Xiaoyu Miao / 蒋镇辉，李佳欣，苗霄宇  
            HKU Business School Shenzhen Research Institute  
            """)
    gr.HTML(
        value= "<br/><p style='margin-top: 1rem, margin-bottom: 1rem'> Please refer to the <a href='https://hkubs.hku.hk/aimodelrankings/report'_target='blank'> report </a> for details on metrics, tasks and models.<br/>Updated 01/2024.</p>")
   


    # 创建一个包含Markdown说明的示例块

    with gr.Tab("🏅 综合榜单（人类裁判）"):
        text_input = None
        text_output = gr.DataFrame(value=show_general_leaderboard(),
                         label='Leaderboard',
                         interactive=False,
                         wrap=True)
        
    with gr.Tab("💬 通用语言能力榜单（人类裁判）"):
        text_input = None
        text_output = gr.DataFrame(value=show_basic_leaderboard(),
                         label='Leaderboard',
                         interactive=False,
                         wrap=True)
    
        
    with gr.Tab("📚 专业学科能力榜单（正确率）"):
        with gr.Tab("学科能力榜单（正确率）"):
            text_input = None
            text_output = gr.DataFrame(value=show_subject_leaderboard(),
                            label='Leaderboard',
                            interactive=False,
                            wrap=True)
        with gr.Tab("中学难度学科能力榜单（正确率）"):
            text_input = None
            text_output = gr.DataFrame(value=show_midschool_leaderboard(),
                            label='Leaderboard',
                            interactive=False,
                            wrap=True)
        with gr.Tab("大学难度学科能力榜单（正确率）"):
            text_input = None
            text_output = gr.DataFrame(value=show_college_leaderboard(),
                            label='Leaderboard',
                            interactive=False,
                            wrap=True)
        
    
    with gr.Tab("⭕️ 安全与责任榜单（人类裁判）"):
        text_input = None
        text_output = gr.DataFrame(value=show_safety_leaderboard(),
                         label='Leaderboard',
                         interactive=False,
                         wrap=True)
    
    gr.Markdown(
            """
            We also employed a fine-tuned GPT-3.5 Turbo as a judge to evaluate large language models through pairwise comparisons. 
            The findings are presented below.
            
            """)
    
    with gr.Tab("🤖 通用语言能力榜单(大模型裁判)"):
        text_input = None
        text_output = gr.DataFrame(value=show_LLM_judge_leaderboard(),
                         label='Leaderboard',
                         interactive=False,
                         wrap=True)
    


demo.launch(share=True)