Installation Issue: 'grain' Dependency Not Found During Gemma Setup​

[pronajit]

I am encountering an issue while attempting to install the Gemma library. During the installation process, I receive the following error message:

ERROR: Could not find a version that satisfies the requirement grain (from gemma) (from versions: none)
ERROR: No matching distribution found for grain
I have verified that my Python environment meets the requirements specified in the pyproject.toml file, including Python version 3.11.9. However, the installation fails due to the missing 'grain' package.​

Upon further investigation, I found that the 'grain' package is not publicly available on PyPI, which may be causing this issue.​
how to resolve this dependency issue? Specifically:
Is there an alternative source or repository where the 'grain' package can be obtained?​
Are there any recommended steps to successfully install Gemma with all its dependencies?​

[Conchylicultor]

Which OS are you using ?

• Grain is published here: https://pypi.org/project/grain/

For supported OS: https://pypi.org/project/grain/

There are some specific CPU architecture in some MaC version for which grain isn't yet supported. Hopefully they will soon.

However, Gemma can be used without Grain though. So it's a bug on us. I'll update the package to allow using Gemma without PyGrain

[pronajit]

Thank you for the clarification and quick response, @Conchylicultor! I'm using Windows 11. I appreciate the update that Gemma can be used without PyGrain and that you'll be updating the package to allow this. I'll proceed with my work using the current workaround and keep an eye out for the update. Thanks again for your support

[Conchylicultor]

Those 2 PRs should skip grain installation on Windows.

• Make grain an optional dependency on Windows #207
• Make grain an optional dependency on Windows google-research/kauldron#1066

It means the data pipeline features won't be available, but you can still call and sample the model with gm.text.ChatSampler().

We'll also need to release new PyPI version of Kauldron & Gemma to push those changes on PyPI.