Apply for a GPU community grant: Academic project

#2
by TimWindecker - opened
Robotic Systems Lab - ETH Zürich org

Hey Hugging Face Team,

We’ve launched a Space for NaviTrace, a novel VQA benchmark designed to evaluate VLMs on embodiment-specific navigation in challenging real-world scenarios. Our work was recently accepted to ICRA 2026, and we’re excited to open up the evaluation to the broader community!

The Goal:
We want this Space to serve as the official public leaderboard where researchers can upload their results and calculate their test scores.

The Bottleneck:
Currently, running on the free CPU tier, a single score calculation takes approximately 2 hours. This high latency makes it difficult for the community to iterate and participate.

The Request:

  • Hardware: Would it be possible to upgrade this Space to a higher-end CPU tier to bring that evaluation time down to a reasonable window?

  • Upload Limits: We also noticed that there is a 1MB file upload limit (linked to the Streamlit AxiosError 413). Since the result files for this benchmark can easily be larger than this, could we have this limit increased for this Space?

You can check out more about our work on our project page. Thanks for the help!

Best,
Tim

Hi @TimWindecker , we've assigned cpu-upgrade to this Space.
I don't think the file upload error is an issue with the Hub. It might be related to Streamlit. At least on a Gradio Space, I can upload larger files without any problems.

Robotic Systems Lab - ETH Zürich org

Thanks, I appreciate the support

TimWindecker changed discussion status to closed

Sign up or log in to comment