README.md

metadata

title: FutureBench Leaderboard
emoji: 🔮
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false

FutureBench Leaderboard App

A minimal Gradio application for viewing FutureBench prediction data. This app downloads datasets from HuggingFace on startup and provides a web interface to explore the data.

Features

• 📊 Data Summary: View dataset statistics and information
• 🔍 Sample Data: Browse sample prediction records
• 📋 About: Learn about the FutureBench system
• 🔄 Auto-refresh: Download latest data on startup
• 📅 Date Range Slider: Filter the leaderboard by a custom date span

Setup

1. Install dependencies:

pip install -r requirements.txt

2. (Optional) Set your HuggingFace token for private repositories:

export HF_TOKEN=your_token_here

Running the App

Launch the Gradio application:

python app.py

The app will:

1. Download datasets from HuggingFace repositories on startup
2. Process the data and create summaries
3. Launch a web interface at http://localhost:7860

Data Sources

The app downloads data from these HuggingFace repositories:

• futurebench/requests - Evaluation queue
• futurebench/results - Evaluation results
• futurebench/data - Main prediction dataset

Structure

• app.py - Main Gradio application
• process_data/ - Data processing utilities
• requirements.txt - Python dependencies
• README.md - This file

Next Steps

This is a minimal version focusing on data download and display. Future enhancements will include:

• Full leaderboard with model rankings
• Interactive filtering and sorting
• Detailed performance metrics
• Model comparison tools