Help with cuda 12.8 for docker

[hakonstoerholt]

I tried installing the standard windows way, but came across an issue with downloading unidic. So I attempted to go the docker way, but I have a new issue here. I am currently running a 5070ti, meaning I need pytorch nightly for cuda 12.8. Anyone know how I might go about solving this?

[DrewThomasson]

Did you try the docker image variant built for cuda 12?

You should be able to ya that by just adding the cuda12 tag at the end of the image name

Like athomasson2/ebook2audiobook:cuda12

Details at this spot in readme -> https://github.com/DrewThomasson/ebook2audiobook?tab=readme-ov-file#docker-gpu-options

[SyamsQ]

Do you make tutorial for adding cuda 12. I still not understand.

[DrewThomasson]

Oh I see that one is built for cuda 121 not 128

Does it have to be the nightly build? Or is cuda128 fine

[hakonstoerholt]

Could you elaborate on what you mean by version number?

I tried again, this time pulling the athomasson2/ebook2audiobook:latest, downloading the newest PyTorch and committing to a new image. Same result, it starts processing using Cuda, but at a certain spot it stops with the error:
XTTS can only generate text with a maximum of 400 tokens..

[hakonstoerholt]

I know it doesn't really fit the original topic anymore, but since I've already mentioned it, I'll continue. I tried running ebook2Audiobook on my Google Colab subscription, and there too did I get the XTTS can only generate text with a maximum of 400 tokens. error.

[DrewThomasson]

Yeah please make a GitHub issue for that,

It doesn't happen with every book and we thought we fixed it last time. :/

[DrewThomasson]

Could you elaborate on what you mean by version number?

I tried again, this time pulling the athomasson2/ebook2audiobook:latest, downloading the newest PyTorch and committing to a new image. Same result, it starts processing using Cuda, but at a certain spot it stops with the error:

XTTS can only generate text with a maximum of 400 tokens..

Well in the gui it should say the version number, but I guess we can confirm this appears in the latest,

Think you could provide a full log of it? And the book causing the error if possible in your new issue report?

Also with the cuda 12.8 I was wondering if you could look at these changes I made to see if I'm making it grab the wrong pip install or something for CUDA 12.8 variant

https://github.com/DrewThomasson/ebook2audiobook/pull/628/files

[SyamsQ]

Where should I put these three files?

[lokmjuhb]

I am using 5070ti as well and running locally, and it could not detect my GPU. As from your discussion, is switching to docker the only way for now?

[lokmjuhb]

it can actually get my gpu

[ROBERT-MCDOWELL]

you must run this gpu test script under the python_env of eb2ab so:
conda activate ./python_env for linux/mac and .\python_env for windows
then run your python command and check the torch package.

[lokmjuhb]

Yes I know. I did (unless I didnt)

[ROBERT-MCDOWELL]

ok so now it needs to install the right torch package into docker... check https://github.com/DrewThomasson/ebook2audiobook?tab=readme-ov-file#docker-gpu-options

[DrewThomasson]

Look at the docker build arguments

You'll have to build the docker image locally if your specifying a torch package
https://github.com/DrewThomasson/ebook2audiobook?tab=readme-ov-file#avalible-docker-build-arguments