app.py

import gradio as gr
import pandas as pd
import numpy as np
import json
import requests


# load from remote json file
FILE_URL = "https://raw.githubusercontent.com/EleutherAI/lm-evaluation-harness/master/results/llama/llama-30B/llama-30B_mmlu_5-shot.json"
response = requests.get(FILE_URL)
data = response.json()

# Create a DataFrame from the results dictionary
data = pd.DataFrame(data['results']).T

# Sort the DataFrame by 'acc' column
data = data.sort_values('acc', ascending=False)

def show_leaderboard():
    # Convert dataframe to html so that it can be displayed properly in Gradio
    return data.to_html()

iface = gr.Interface(fn=show_leaderboard, inputs=[], outputs="html")

# Run the interface. 
# Note: you don't need to use .launch() in Hugging Face Spaces, this is for local testing.
iface.launch()