Make Embodied First Exam an official benchmark on the Hub?

#2
by SaylorTwift - opened

Hi @JamesK2W 👋

I'm Nathan from Hugging Face. I lead our "Leaderboards on the Hub" effort. Congrats on Embodied First Exam, it's been going around a lot internally!

Since the tasks already live here and it runs on Harbor, we'd love to help make it an official benchmark on the Hub:

  • You add an eval.yaml to this dataset (a few lines: name, description, one task per setting, e.g. privileged / standard / real-time solve rate), and the Hub shows a leaderboard on this page.
  • Each model's score also appears on its model page, linking back to your benchmark. You stay in control of what's listed (you can hide results you dispute).
  • We can seed it with the open-weight runs from your leaderboard (Kimi K3.1 Pro, DeepSeek V4.1 Flash, GLM-5.3 Flash) and do joint comms at launch.

Docs: https://huggingface.co/docs/hub/eval-results · https://huggingface.co/docs/hub/leaderboard-data-guide

Happy to open the eval.yaml PR for you to review. Which metric would you want front and center?

Thanks!
Nathan (HF @SaylorTwift · GitHub @NathanHB )

Sign up or log in to comment