app.py

import gradio as gr
import pandas as pd
import os
from huggingface_hub import snapshot_download
from apscheduler.schedulers.background import BackgroundScheduler

from src.display.about import (
    CITATION_BUTTON_LABEL,
    CITATION_BUTTON_TEXT,
    EVALUATION_QUEUE_TEXT,
    INTRODUCTION_TEXT,
    LLM_BENCHMARKS_TEXT,
    TITLE,
)
from src.display.css_html_js import custom_css
from src.envs import API

# clone / pull the lmeh eval data
TOKEN = os.environ.get("TOKEN", None)
RESULTS_REPO = f"SeaLLMs/SeaExam-results"
CACHE_PATH=os.getenv("HF_HOME", ".")
EVAL_RESULTS_PATH = os.path.join(CACHE_PATH, "eval-results")
print(EVAL_RESULTS_PATH)
snapshot_download(
    repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", 
    token=TOKEN
)

def restart_space():
    API.restart_space(repo_id="SeaLLMs/SeaExam_leaderboard", token=TOKEN)

# Load the CSV file
# def load_csv(file_path):
#     data = pd.read_csv(file_path)
#     return data

def load_data(data_path):
    df = pd.read_csv(data_path, skiprows=1, header=0).dropna()

    columns = ['Model', 'type', 'open?', 'shot', 'en', 'zh', 'id', 'th', 'vi', 'avg', 'avg_sea']
    

    # Splitting into three separate DataFrames based on the groups M3Exam and MMLU and average
    df_m3exam = df.iloc[:, :11]  # M3Exam columns
    df_mmlu = df.iloc[:, [0, 1, 2, 3, 11, 12, 13, 14, 15, 16, 17]]  # MMLU columns
    df_avg = df.iloc[:, [0, 1, 2, 3, 18, 19, 20, 21, 22, 23, 24]]  # Average columns
    df_mmlu.columns = columns
    df_avg.columns = columns

    # # multiply the values in the ['en', 'zh', 'id', 'th', 'vi', 'avg', 'avg_sea'] by 100 and display as 1 decimal
    for df_tmp in [df_m3exam, df_mmlu, df_avg]:
        df_tmp[['en', 'zh', 'id', 'th', 'vi', 'avg', 'avg_sea']] *= 100
        df_tmp[['en', 'zh', 'id', 'th', 'vi', 'avg', 'avg_sea']] = df_tmp[['en', 'zh', 'id', 'th', 'vi', 'avg', 'avg_sea']].round(2)
        df_tmp['rank'] = df_tmp['avg'].rank(ascending=False).astype(int)
    # change the order of the columns to ['Model', 'type', 'open?', 'shot', 'avg', 'avg_sea', 'en', 'zh', 'id', 'th', 'vi'] 
    # and sort the columns by 'avg' in descending order
    columns_sorted = ['rank','Model', 'type', 'open?', 'shot', 'avg', 'avg_sea', 'en', 'zh', 'id', 'th', 'vi']
    df_m3exam = df_m3exam[columns_sorted].sort_values(by='avg', ascending=False)
    df_mmlu = df_mmlu[columns_sorted].sort_values(by='avg', ascending=False)
    df_avg = df_avg[columns_sorted].sort_values(by='avg', ascending=False)

    # change the column name from 'avg' to 'avg⬆️'
    df_m3exam = df_m3exam.rename(columns={'avg': 'avg⬆️'})
    df_mmlu = df_mmlu.rename(columns={'avg': 'avg⬆️'})
    df_avg = df_avg.rename(columns={'avg': 'avg⬆️'})
    
    return df_m3exam, df_mmlu, df_avg

# Example path to your CSV file
csv_path = f'{EVAL_RESULTS_PATH}/SeaExam_results.csv'
# data = load_csv(csv_path)

df_m3exam, df_mmlu, df_avg = load_data(csv_path)

# def show_data():
#     return data

# iface = gr.Interface(fn=show_data, inputs = None, outputs="dataframe", title="SeaExam Leaderboard", 
#                      description="Leaderboard for the SeaExam competition.")
# iface.launch()

demo = gr.Blocks(css=custom_css)
with demo:
    gr.HTML(TITLE)
    # gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem("🏅 Summary", elem_id="llm-benchmark-Sum", id=0):
            leaderboard_table = gr.components.Dataframe(
                value=df_avg,
                # value=leaderboard_df[
                #     [c.name for c in fields(AutoEvalColumn) if c.never_hidden]
                #     + shown_columns.value
                #     + [AutoEvalColumn.dummy.name]
                # ],
                # headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value,
                # datatype=TYPES,
                # elem_id="leaderboard-table",
                interactive=False,
                visible=True,
                # column_widths=["20%", "6%", "8%", "6%", "8%", "8%", "6%", "6%", "6%", "6%", "6%"],
            )
        with gr.TabItem("🏅 M3Exam", elem_id="llm-benchmark-M3Exam", id=1):
            leaderboard_table = gr.components.Dataframe(
                value=df_m3exam,
                interactive=False,
                visible=True,
            )
        with gr.TabItem("🏅 MMLU", elem_id="llm-benchmark-MMLU", id=2):
            leaderboard_table = gr.components.Dataframe(
                value=df_mmlu,
                interactive=False,
                visible=True,
            )
        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=3):
            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")

demo.launch()

scheduler = BackgroundScheduler()
scheduler.add_job(restart_space, "interval", seconds=1800)
scheduler.start()
demo.queue(default_concurrency_limit=40).launch(share=True)