DREAMBOOTH CUDA SETUP: Setup Failed!

[k0xxxx]

Hello, please tell me, I am launching on google colab https://colab.research.google.com/github/Linaqruf/kohya-trainer/blob/main/kohya-LoRA-dreambooth.ipynb#scrollTo=p_SHtbFwHVl1, but at the training launch stages A lot of errors appear, although just yesterday everything was working. Any ideas how to fix it?

[toshimickey]

samme issue here

[PlumButa]

Same issue 😢 I noticed that the following error message appears during step "1.1. Install Dependencies":

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behavior is the source of the following dependency conflicts. pandas-gbq 0.19.2 requires google-auth-oauthlib>=0.7.0, but you have google-auth-oauthlib 0.4.6 which is incompatible.

Could it be related to this error message?

[DEX-1101]

CUDA backend failed to initialize: Found CUDA version 12010, but JAX was built against version 12020, which is newer. The copy of CUDA that is installed must be at least as new as the version against which JAX was built. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.

the error is caused by CUDA, the other repo is also same getting same error

[Omenizer]

Same issue, error in the dependencies, and when training starts I'm getting the CUDA error

[chelseasn0210]

I'm having the same issue

[SkyProlk2022]

I'm having the same issue too

[kerlina1939]

Maybe this method can be used https://github.com/hollowstrawberry/kohya-colab/issues/69 It seems to be effective, but I don’t know where to add !pip install --upgrade bitsandbytes . If anyone knows how to use it, please take a screenshot, I would like to know how to use it. Thanks! !

[Atheuz]

Create a new cell and execute this:

!wget https://raw.githubusercontent.com/TimDettmers/bitsandbytes/main/install_cuda.sh

Then create another new cell and execute this:

!bash install_cuda.sh 118  /usr/local/ 0

That worked for me.

The problem appears to be that CUDA 11.8 is not installed, but 12.2 is, the colab notebook expects 11.8. The above installs CUDA 11.8 again.

Though this feels like a bandaid, the other fix seems better, but I haven't tried it.

[wesleywil]

Create a new cell and execute this:

!wget https://raw.githubusercontent.com/TimDettmers/bitsandbytes/main/install_cuda.sh

Then create another new cell and execute this:

!bash install_cuda.sh 118  /usr/local/ 0

That worked for me.

The problem appears to be that CUDA 11.8 is not installed, but 12.2 is, the colab notebook expects 11.8. The above installs CUDA 11.8 again.

This work for me as well, thanks man

[chelseasn0210]

Create a new cell and execute this:

!wget https://raw.githubusercontent.com/TimDettmers/bitsandbytes/main/install_cuda.sh

Then create another new cell and execute this:

!bash install_cuda.sh 118  /usr/local/ 0

That worked for me.

The problem appears to be that CUDA 11.8 is not installed, but 12.2 is, the colab notebook expects 11.8. The above installs CUDA 11.8 again.

Though this feels like a bandaid, the other fix seems better, but I haven't tried it.

Is this done before mounting to Google Drive? Or is this done after?

[SkyProlk2022]

Create a new cell and execute this:

!wget https://raw.githubusercontent.com/TimDettmers/bitsandbytes/main/install_cuda.sh

Then create another new cell and execute this:

!bash install_cuda.sh 118  /usr/local/ 0

That worked for me.

The problem appears to be that CUDA 11.8 is not installed, but 12.2 is, the colab notebook expects 11.8. The above installs CUDA 11.8 again.

Though this feels like a bandaid, the other fix seems better, but I haven't tried it.

Thank you sir!! It's working

[SkyProlk2022]

Create a new cell and execute this:

!wget https://raw.githubusercontent.com/TimDettmers/bitsandbytes/main/install_cuda.sh

Then create another new cell and execute this:

!bash install_cuda.sh 118  /usr/local/ 0

That worked for me. The problem appears to be that CUDA 11.8 is not installed, but 12.2 is, the colab notebook expects 11.8. The above installs CUDA 11.8 again. Though this feels like a bandaid, the other fix seems better, but I haven't tried it.

Is this done before mounting to Google Drive? Or is this done after?

It worked for me when I did it before

[chelseasn0210]

Create a new cell and execute this:

!wget https://raw.githubusercontent.com/TimDettmers/bitsandbytes/main/install_cuda.sh

Then create another new cell and execute this:

!bash install_cuda.sh 118  /usr/local/ 0

That worked for me.

The problem appears to be that CUDA 11.8 is not installed, but 12.2 is, the colab notebook expects 11.8. The above installs CUDA 11.8 again.

Though this feels like a bandaid, the other fix seems better, but I haven't tried it.

Is this done before mounting to Google Drive? Or is this done after?

It worked for me when I did it before

It worked!!! Thank you so much 🙌

[Linaqruf]

current state explained here, thank you for waiting https://github.com/Linaqruf/kohya-trainer/issues/321#issuecomment-1861318632 let me know if it's still throw error

[MajorGruberth]

I still have the Coda-problem: I installed the 2 extra cells with de-install CUDA 12.2 and install 11.8, but the notebook still ceased to work with the same error at 4.2.1 I pushed the 2 extra cells on top, didn't work, I pushed them behind the dependencies cell - didn't work. How come?

[Omenizer]

You don't need the cells, the bug has been fixed

[MajorGruberth]

Well, I tried a fresh notebook kohya-LoRA-dreambooth.ipynb and found the good old error again: "Cuda backend failed to initialize...found cuBLAS version 120103..."

[Omenizer]

Did it work?

[MajorGruberth]

Nope... see above. Same error :-(

[Omenizer]

I'm asking because the author wrote that the error might still appear but the notebook would work, which is how it has been for me since his fix

[MajorGruberth]

Then I am wondering, why it's not working here...? Same platform, same notebook - different results?