Spaces:

Duplicated from  hysts/LoRA-SD-training

nupurkmr9

/

custom-diffusion

like 109

Build error

Hi, are you running the space locally or on huggingface spaces? And which GPU?

I am running it on huggingface spaces using T4 small GPU . I can see my results/delta.bin file . Before that I tried the other bin files but nothing is showing up.

I am running the demo on A10G and don't face any issues. But have observed this sometimes on a T4. Still trying to debug the cause.

Thanks, as comparison similar workloads such as training and inference on diffusers work perfectly fine on Google Colab using a T4 without any issues. Which is similar to the T4 I used here on huggingface spaces. I will try to port this over to Colab or maybe someone would beat me to it.

Thanks. I think the defaults params during inference are the issue. With smaller batch-size and less number of steps it should work.

Thanks @nupurkmr9 , it worked if I used batch size of 1