No matching distribution found for nvidia-cudnn-cu12

[cardonymous]

Another dependeny issue arises on my laptop: “No matching distribution found for nvidia-cudnn-cu12==8.9.2.26”

Issue occurs with python 3.11.9 on laptop with Windows 10, Intel Core i5 8th Gen and Nvidia geforce 1050. In the past, all attemps for cuda installations have made trouble and failed. Is cuda required or would it be possible to make it optional?

[labrunhosarodrigues]

CUDA is not required, it just speeds up CLIP related inference tasks. It is another problem with full requirement.txt dumps. Please try and remove all cuda related dependencies.

[cardonymous]

I made some progress on the dependencies. However, I received another "no matching distribution" error for sklearn-0.0.post11. Aditionally, there's a warning that sklearn is deprecated and one should use scikit-learn instead. I took sklearn out since scikit-learn is already installed.

Package tensorflow-io-gcs-filesystem also fails since it's missing a windows build. Removed that one, too.

Next I have an error for triton==2.1.0. What is unusual is that usually pip tries different versions of a package and lists them all. However, for this one, pip seems to find no version at all.

Could not find a version that satisfies the requirement triton==2.1.0 (from versions: none)
No matching distribution found for triton==2.1.0 

[labrunhosarodrigues]

It seems that triton only supports Linux: https://github.com/triton-lang/triton?tab=readme-ov-file#compatibility I believe that is another instance of dependencies of dependencies that the code can run without.

[cardonymous]

[labrunhosarodrigues]

None of those dependencies need to be explicitly installed (dependencies of dependencies). But I also see that you are using python 3.11, and the repository was created using python 3.8, which explains some of those incompatibilities

[cardonymous]