Spaces:
Running
on
Zero
Apply for community grant: Academic project (gpu and storage)
This is the official implementation demo of the arXiv paper: Diffusion As Shader (arxiv.org/abs/2501.03847). We are excited to share our work with the broader community and demonstrate the capabilities of our model in real-time applications. In response to the invitation from Niels at Hugging Face (invitation link: github.com/IGL-HKUST/DiffusionAsShader/issues/2), we are applying for a ZeroGPU grant to host our interactive demo on HF Spaces. The ZeroGPU grant will allow us to leverage A100 GPUs to create a seamless and efficient demo, showcasing the model's ability to generate high-quality shader effects in real-time. This will not only enhance the discoverability of our work but also foster collaboration and innovation within the community.
Hi @EXCAI , we've assigned ZeroGPU to this Space. Please check the compatibility and usage sections of this page so your Space can run on ZeroGPU.
Hi
@hysts
! Thanks a lot for your help! But we are still a little confused about the configuration of ZeroGPU.
When running applications on ZeroGPU, we're encountering an issue where cupy.cuda.get_cuda_path() returns None, causing a TypeError when trying to set it as an environment variable. This happens because the CUDA installation path cannot be automatically detected in this command: os.environ['CUDA_HOME'] = cupy.cuda.get_cuda_path() . So we'd like to know the correct CUDA installation path on ZeroGPU servers. Or is there a recommended way to handle CUDA path detection when using ZeroGPU? This information would help us properly configure our application to work reliably in the ZeroGPU environment.
@EXCAI
Thanks for checking if your Space works on ZeroGPU. In ZeroGPU Spaces, CUDA is only available inside functions decorated with @spaces.GPU. If the cupy package requires CUDA at import time, it might not be compatible with ZeroGPU. I'm not sure if there's a workaround, so let me cc
@cbensimon
for more details.