Unable to replicate the results
psurya1994 opened this issue · 3 comments
I attempted to replicate the results shown in the README by using the codebase (training with the config in config/sample.yml), but was unable to do so. I've attached the sample/step_400.gif generated below for this training run.
Can you help me understand what I might be missing?
In personal communication with the repo owner, I was told it was probably an issue with how I'm loading weights (I did make a couple of changes to my weights part of the code). Will take a closer look to ensure the weights are being loaded correctly. Closing the issue for now. Will reopen in case I still face issues.
My issue has been resolved. Thanks to @bryandlee !
The issue in my case was that I was using directly using the hugging face path for my model, instead of saving it locally (like the code does). Specifically:
pretrained_model_path: runwayml/stable-diffusion-v1-5. Because of which, the model was not loading the weights for the UNet correctly.
I fixed the issue by appropriately changing the path on this line of the code.
Hi @psurya1994,
I am also trying to save the model weights locally. However, I am not that familiar with stable diffusion, so I wonder how should I prepare or get the weights to run the code.
Thanks in advance!
Junting