RuntimeError: CUDA out of memory. Tried to allocate 114.00 MiB about dreambooth-stable-diffusion HOT 11 CLOSED

I've been having the same issue. I've followed the steps multiple times trying both 3090 and A5000 on the regular and community areas, but nothing works.

from dreambooth-stable-diffusion.

I checked my dataset and I had one image that was 511x512. I fixed it but I still get this error. I want to note that this dataset worked perfectly fine yesterday on the previous commits so its something in the new changes that broke something.

from dreambooth-stable-diffusion.

Not for me. i stayed on Stable diffusion,
I assure myself that i had wrote 60 GB disk and 60 GB Volume and after i run. After, before opening the Joepenna Run pod. You need to open Terminal and kills two cells.
In Terminal open a new tab, and type "ps aux" to find the PID for relauncher and webui, kill both process and restart the training. Otherwise your trainig will be killed. And after, train normally.

from dreambooth-stable-diffusion.

I've something for you. In my case it was because one of my photo aren't in 512 x 512 after checking. In resize it and it's done. But now... my training is killed so... Solve an issue and you've an other one ahah

from dreambooth-stable-diffusion.

Hum, i restart a pod on Runpod because i think i do not allowed 60 GB Disk and 60 Gb Volume. When u changed Pytorch to Stable Diff, its reset. I retry it, make the changes and it was okay for me

from dreambooth-stable-diffusion.

But my training was killed after. It is happen to you, you have a the correction on issues section. Sorry for my ENglish

from dreambooth-stable-diffusion.

When u changed Pytorch to Stable Diff, its reset.

Not sure I understand. Using a different version of pytorch made your CUDA memory error go away?

from dreambooth-stable-diffusion.

Oh Sorry, you don't understand. So before you lunch note pad u need to check if your GB disk and GB volume are enough. Per default it is 20/20. You need to have 60/60 on RunPod.

from dreambooth-stable-diffusion.

Is the pod GPU memory usage at 0% on the runpod website under my pods? For me, it wasn't. It was at a few % usage after installing the dependencies (something didn't get cleaned up properly?). Restarting the pod fixed the issue. Make sure to run the part that has the dataset= in it, to properly set the variable for training - or create a new cell and set that variable.

from dreambooth-stable-diffusion.

Restart the notebook. Then, try to run things.

from dreambooth-stable-diffusion.

You need to open Terminal and kills two cells.
In Terminal open a new tab, and type "ps aux" to find the PID for relauncher and webui, kill both process and restart the training. Otherwise your trainig will be killed. And after, train normally.

^this

See also:

• #95

For those on runpod either do what's suggested above or connect to a web terminal and kill those two processes:

sh -c python webui.py --port 3000 --ckpt /workspace/stable-diffusion-webui/v1-5-pruned-emaonl...
python webui.py --port 3000 --ckpt /workspace/stable-diffusion-webui/v1-5-pruned-emaonly.ckpt...

Originally posted by @zoellner in #95 (comment)

from dreambooth-stable-diffusion.