Blocked while training attempt - name 'str2optimizer8bit_blockwise' is not defined" #382

Hey there! I'm trying to train a new voice, and I followed everything in the wiki, but when I hit the "Train" button, I have this message after a few seconds:

"[Training] [2023-09-13T22:40:31.710038] NameError: name 'str2optimizer8bit_blockwise' is not defined"

I'm on Ubuntu 22.04 LTS, CUDA 12.2. Anyone has an idea?

Thank you

Hey there! I'm trying to train a new voice, and I followed everything in the wiki, but when I hit the "Train" button, I have this message after a few seconds: "[Training] [2023-09-13T22:40:31.710038] NameError: name 'str2optimizer8bit_blockwise' is not defined" I'm on Ubuntu 22.04 LTS, CUDA 12.2. Anyone has an idea? Thank you

According to https://github.com/TimDettmers/bitsandbytes/issues/62 you should have a bunch of messages from BitsAndBytes hinting at a misconfigured system.

The solution seems to be passing in your path to your CUDA libs to LD_LIBRARY_PATH.

According to https://github.com/TimDettmers/bitsandbytes/issues/62 you should have a bunch of messages from BitsAndBytes hinting at a misconfigured system. The solution seems to be passing in your path to your CUDA libs to LD_LIBRARY_PATH.

Hey there! Thank you for your quick answer :) I'll investigate this and let you know!

Have a great rest of the day/night :)

Hey there! Thank you for your quick answer :) I'll investigate this and let you know! Have a great rest of the day/night :)

Hey again! So the training has been running for about 8 hours now so I think it's safe to assume that the issue is fixed ^^What I did basically is reinstall everyhing by following the instructions of the wiki (under the "Setup" section of the Installation page, I didn't follow the instructions for Bark and Vall-E):

Recreate a Conda environment built on Python 3.10
Cloned the repo
Launched the setup cuda.sh
And there it worked. I went to check the environment variables once everything was done, and indeed the LD_LIBRARY_VARIABLE was set to the cuda installation of the environment, so I guess it was indeed that!

Thank you again for the lead :)

Hey again! So the training has been running for about 8 hours now so I think it's safe to assume that the issue is fixed ^^What I did basically is reinstall everyhing by following the instructions of the wiki (under the "Setup" section of the Installation page, I didn't follow the instructions for Bark and Vall-E): - Recreate a Conda environment built on Python 3.10 - Cloned the repo - Launched the setup cuda.sh And there it worked. I went to check the environment variables once everything was done, and indeed the LD_LIBRARY_VARIABLE was set to the cuda installation of the environment, so I guess it was indeed that! Thank you again for the lead :)

Labels Milestones

Blocked while training attempt - name 'str2optimizer8bit_blockwise' is not defined" #382