app.py

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

# URLs for the two JSON files
FILE_URL_1 = "https://raw.githubusercontent.com/EleutherAI/lm-evaluation-harness/master/results/llama/llama-30B/llama-30B_mmlu_5-shot.json"
FILE_URL_2 = "https://raw.githubusercontent.com/EleutherAI/lm-evaluation-harness/master/results/llama/llama-13B/llama-13B_mmlu_5-shot.json"

# Load data from both URLs
response1 = requests.get(FILE_URL_1)
data1 = response1.json()
response2 = requests.get(FILE_URL_2)
data2 = response2.json()

# Convert data from both URLs into DataFrames
data1_df = pd.DataFrame(data1['results']).T
data2_df = pd.DataFrame(data2['results']).T

# Rename 'acc' column to respective file names
data1_df = data1_df.rename(columns={'acc': 'acc_file_1'})
data2_df = data2_df.rename(columns={'acc': 'acc_file_2'})

# Merge the dataframes on index (Here index is the sub-test names)
data = pd.merge(data1_df['acc_file_1'], data2_df['acc_file_2'], left_index=True, right_index=True)

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()