README.md

metadata

title: Arabic LLM Leaderboard
emoji: 🏆
colorFrom: green
colorTo: yellow
sdk: docker
app_port: 7860
pinned: true
license: apache-2.0
short_description: Benchmarking LLMs on Modern Standard Arabic
hf_oauth: true
hf_oauth_expiration_minutes: 480
tags:
  - leaderboard
  - arabic
  - llm
  - evaluation

Arabic LLM Leaderboard

A single-page leaderboard for Arabic (Modern Standard Arabic) LLMs. It reads results live (client-side) from the Mushari440/results dataset and the evaluation queue from Mushari440/requests.

Each model is scored on 13 capability & domain dimensions; the headline Average is the item-weighted mean (every one of the 29,417 benchmark items counts equally).

The page is a single self-contained index.html (vanilla JS) served by a small FastAPI app. Anyone can submit their own model: sign in with your Hugging Face account, fill the form, and the app validates the model on the Hub and enqueues it (PENDING) into Mushari440/requests. The write is done with the Space's own token (a private Space secret) — your token is never used to write and never leaves your browser; sign-in only records your username on the request.