Spaces:
Sleeping
Apply for community grant: Academic project (gpu)
Smooth Diffusion is a new category of diffusion models that is simultaneously high-performing and smooth. We want to apply for a GPU grant.
Hi @JiayiGuo821 , 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 for the ZeroGPU grant. Smooth Diffusion is successfully running on ZeroGPU. We have tested several easy cases. It works.
However, when running cases that consume more GPU resources and inference time, errors like gradio.exceptions.Error: 'GPU task aborted' sometimes raised. Is that expected on ZeroGPU?
@JiayiGuo821
Thanks for testing ZeroGPU!
The error is raised when the function decorated with @spaces.GPU takes longer than the time limit, which is 60 seconds by default. You can change it using the duration parameter, like @spaces.GPU(duration=120).
FYI, this parameter is supposed to specify the expected maximum inference time for the function. On ZeroGPU, each user has a few minutes worth of GPU quota per a few hours, and this parameter is used to check if the user can still have enough quota to run this function. For example, a user with 90 seconds quota cannot run a function with duration=120 and they have to wait for their quota to be recharged.
@hysts
Thanks for the clarification!
Is that possible for Smooth Diffusion to get a GPU grant? Some cases typically require an inference time of 3-5 minutes.
Smooth Diffusion has earned 200+ stars on GitHub. We believe it first explores the crucial latent space smoothness of diffusion models and will inspire future research.
@JiayiGuo821 Ah, I see. In that case, we can assign A10G. (A100 is not available for grants except for ZeroGPU). I'll switch the hardware to a10g-small, then.
@hysts thank you!
I switched the hardware. The @spaces.GPU decorator does nothing on a normal GPU, so you don't have to change your code.