Spaces:
Runtime error
Apply for community grant: Academic project (gpu)
We would like to offer a demo for our research project, which is an upgraded version of OV-Seg. The demo of OV-Seg however requires GPU. Is it possible to get academic credit for a GPU tier?
@hysts Very good point, we just wanted to wait for the weekend arxiv digest, but you can find our preprint here now: https://arxiv.org/abs/2405.20141
@blumh
Awesome! Thanks for the link! I just assigned ZeroGPU (docs) to this Space.
It seems that the OV-Seg Space is using detectron2 and I'm not sure if it's compatible with ZeroGPU, though. Some libraries, like mmcv, are not compatible with Zero. So, just let us know if your Space doesn't work properly on Zero. We can still assign a normal GPU like L4 as a grant with a shorter sleep time.
FYI, we recently started using ZeroGPU as the default hardware for grants because it would reduce our overall infra cost in the long run and would improve the UX with longer sleep time, parallel execution, etc., so we'd appreciate it if you could check if your Space can run on Zero first.
Thanks a lot, I will check it out and let you know if this works!
@hysts so I just checked compatibitlies, it seems ZeroGPU requires pytorch >=2.0 while the newest pytorch supported by detectron2 is 1.10: https://detectron2.readthedocs.io/en/latest/tutorials/install.html
@blumh Ohhh, I didn't know that. Thanks for checking! I kind of feel that it may still work with torch 2.x even though the documentation is not updated because I think torch is mostly backward compatible and doesn't have much breaking changes when updated to v2. That being said, I guess I'll just assign a normal GPU to this Space for now.
I just assigned t4-medium as it's the hardware assigned to the original OV-Seg Space.