Please fix the notebook, it refuses to work. Install Requirements tab

[VladAdushev]

Describe the bug

Please fix the notebook, it refuses to work. Install Requirements tab

Reproduction

Please fix the notebook, it refuses to work. Install Requirements tab

Logs

No response

System Info

windows 11

[VladAdushev]

https://github.com/ShivamShrirao/diffusers/tree/main/examples/dreambooth

[isMiaArt]

I have tried different ways but have not found the solution to the problem. I hope that soon the author will be able to fix it

[isMiaArt]

After trying different solutions, so far the only thing that has worked for me is this configuration.

Maybe it can also help those of you who have the same problem temporarily until the author updates the notebook.

I share the added code for convenience

%pip install torch==2.0.0+cu118 torchvision torchaudio torchdata torchtext --index-url https://download.pytorch.org/whl/cu118

import torchvision print(torchvision.version)

%pip install triton==2.0.0

[VladAdushev]

Thank you. This method is working. Yes, there are nuances. Gives an error, but continues to work. Finally makes a mistake on the "Inference" tab. But the model creates what was actually required.

[kamikariat]

Thanks @isMiaArt - worked!

[kevindmccoy]

Thanks @isMiaArt - it seems to be working for me in Google Colab also! I saved a copy of ShivamShrirao's notebook to my Google Drive... it hasn't made it to Running Training for me in days until now!

[alexeyugn]

Have you guys tried the created model after those errors? Because mine was created "well" too but crushes the SD while generating using it.

[kamikariat]

@alexeyugn What do you mean by create model?

Are you referring to training or the inference part.

[VladAdushev]

Have you guys tried the created model after those errors? Because mine was created "well" too but crushes the SD while generating using it.

I confirm that this model does not work with automatics 111. I run it through the SDNext shell. https://github.com/Em1tSan/stable-diffusion-portable

[alexeyugn]

@alexeyugn What do you mean by create model? Are you referring to training or the inference part.

Both. I trained the model and .ckpt was created. Downloaded to SD Automatic1111, it crashed. Then tried to generate images right here, crashed also.

[kevindmccoy]

I'm using NMKD, it works after merging the new model 95% with sd-1.5 5%. NMKD v1.11 has a problem with models created since about 7/24/2023 - NMKD v1.9.1 does work with them without merging though

[VladAdushev]

Shivam, please fix the notebook. If there is no way to fix it, write so, we will look for other options.

[chchchadzilla]

I have tried different ways but have not found the solution to the problem. I hope that soon the author will be able to fix it

this doesn't fix it all, but it'll at least get it to work:

1) add new code box after installation 2) add this code:

# prompt: This will fix the colab book enough to use it-- sorry if you still get error messages. At least training should complete.

!pip uninstall xformers torch torchvision -y
!pip install torch==2.1.0 torchvision xformers -q

like i said it doesn't fix it, but it'll let you run training at least. cheers.

[chchchadzilla]

Shivam, please fix the notebook. If there is no way to fix it, write so, we will look for other options.

see my reply above.

[VladAdushev]

see my reply above.

I've looked at your answer, even used it, but it's not a full-fledged fix. Not all SD versions can read these ckpt.

[chchchadzilla]

see my reply above.

I've looked at your answer, even used it, but it's not a full-fledged fix. Not all SD versions can read these ckpt.

Right. Which is exactly what I said.

But, without diving head-first into train_dreambooth.py and figuring it out for yourself, and since this colab is a super convenient (i.e. already learned) way to train models that we all still obviously use, the fact is that you can train a functional diffusers weights/automatic1111 ckpt file that will work just fine using the steps I described.

You can then use one of the half-dozen-or-so auto1111 ext that converts ckpt to safetensors, or use kohya to extract the lora, and bingo bango. Dishes are done, dude.

[VladAdushev]

Who knows, tell me or give me a link. Is it possible to use this collage on a home computer (the video card allows). Apparently the authors of this creation scored on their product...