Spaces:
Running on Zero
Apply for community grant: Academic project (gpu)
Hi Hugging Face team, I would like to apply for a ZeroGPU community grant for our Space. Our project is Pixal3D: Pixel-Aligned 3D Generation from Images, an academic image-to-3D generation project that has been accepted by SIGGRAPH 2026.
This Space is intended to provide a public demo for the research community. Since image-to-3D inference requires GPU acceleration, ZeroGPU would help users try the model directly in the browser more smoothly. Thank you for considering our request.
Hi @Yang2001 , we've assigned ZeroGPU to this Space. Please check the compatibility and usage sections of this page so your Space can run on ZeroGPU.
If you can, we ask that you upgrade to Enterprise to enjoy higher ZeroGPU quota and other features like Dev Mode, Private Storage, and more: hf.co/enterprise
Hi @hysts ,
Thank you very much for providing ZeroGPU access for our Space. We really appreciate the support.
However, we noticed that the ZeroGPU backend now seems to be using an RTX PRO 6000 instead of an H200. At the same time, the minimum required PyTorch version also appears to have changed. As a result, our Space can no longer start properly.
The main issue is that our Space depends on FlashAttention 3, which is closely tied to the GPU backend and supported PyTorch/CUDA versions. Could you please help confirm the current recommended configuration for ZeroGPU in this case? Specifically, we would like to know which GPU backend, CUDA version, PyTorch version, and FlashAttention version we should target to make the Space work correctly.
Thank you again for your help and support.
Best regards,
Dongyang
Hi @Yang2001 , sorry about the sudden hardware change. I checked internally, and it seems the ZeroGPU hardware was changed from H200 to RTX PRO 6000 (Blackwell, sm_120) due to availability constraints on our infra side.
The docs haven't been updated yet, but here are the currently supported versions on the new environment:
- PyTorch: one of 2.8.0, 2.9.1, 2.10.0, or 2.11.0 (note: 2.9.0 is not supported)
- CUDA: 13.0
- Python: 3.10.13 (default) or 3.12.12. Python 3.10 reaches EOL in October 2026, so this might be a good opportunity to switch to python_version: "3.12".
For testing your build on the new hardware, please create a private Space and share the URL. I'll assign ZeroGPU to it so you can verify there.
Hi @hysts ,
Thank you very much for the detailed clarification and for checking internally.
We have now finished debugging and updated our Space to work with the new ZeroGPU environment. Could you please help switch/reassign our Space back to ZeroGPU when convenient?
Once ZeroGPU is ready, we will restart the demo and verify that everything is working properly.
Thanks again for your help and support.
Best regards,
Dongyang