Automatic1111 Dreambooth extension suddenly OOM #308
Replies: 8 comments 10 replies
-
Yes, Dreambooth extension is broken after the huge update, OOMs on all possible options. |
Beta Was this translation helpful? Give feedback.
-
SO, rolling back doesn't help anybody, really. I understand your frustration, but as a developer, I can't fix a problem with the current software if people aren't using it. That said, I've pushed a new update. The default parameters for a model aren't necessarily for everybody, but there's a performance wizard button that should help get you close to the right settings to not OOM. |
Beta Was this translation helpful? Give feedback.
-
I am having similar issues. |
Beta Was this translation helpful? Give feedback.
-
It's worth updating to today's version, there have been many optimizations. |
Beta Was this translation helpful? Give feedback.
-
Same here. I get "IndexError: tuple index out of range". Updated a few minutes ago. But no change. |
Beta Was this translation helpful? Give feedback.
-
I updated Dreambooth to 1d7d341. |
Beta Was this translation helpful? Give feedback.
-
In the latest version I am getting the same error. But this time it said loading dreambooth when I pressed Train. But none of the iterations incremented and just showed 0/10000 and finished. It obviously didn't train anything. Not sure if this issue is known. |
Beta Was this translation helpful? Give feedback.
-
I have 24gb vram. |
Beta Was this translation helpful? Give feedback.
-
I have a RTX 2090 Ti with 12GB VRAM. I got the Dreambooth extension to work almost as soon as it was first released. I was able to train several checkpoints and I was satisfied with the results.
Now it doesn't work anymore. I think something updated and it broke. Whereas before I was able to train in Dreambooth as much as I wanted, now it just plain won't. OOM. I didn't change anything on my computer. I had configured the webui-user.bat to automatically update A1111 and the extension each time I started it. Never doing that again.
The OOM error doesn't happen when I try to train in Dreambooth. Ever since last night I haven't been able to get that far! Now whenever I try to CREATE a model, I get this error:
RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 11.00 GiB total capacity; 10.06 GiB already allocated; 0 bytes free; 10.26 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
I never had this error before. I swear I changed nothing other than allow A1111 and Dreambooth to update themself on their own.
So I have spent all of this morning wiping my workstation, re-installing Windows 10, and without installing anything else I have re-installed Python, Git, Pytorch, CUDA, etcetera ad nauseam infinitum. Automatic1111 works fine. Dreambooth doesn't.
Any help is appreciated.
Beta Was this translation helpful? Give feedback.
All reactions