google / gemma.cpp

lightweight, standalone C++ inference engine for Google's Gemma models.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Seeking feedbacks for python wrapper of gemma.cpp

namtranase opened this issue · comments

Hello, gemma.cpp contributors and community!

I want to express my sincere gratitude for developing gemma.cpp. This library has been a significant learning resource for me.

I've embarked on creating a Python wrapper for gemma.cpp to further my understanding and apply the techniques you've incorporated into the library. I started with libgemma guidelines and used pybind11 for the initial wrapping. The project is in its early stages, and I would greatly value any suggestions for improvement.

Thank you once again to the entire team for your pioneering efforts.

You can find my repository here: gemma-cpp-python

Hi and thanks for jumping on this quickly!

Something I'd be curious on is what would you (others) be interested in using the binding for? How your project is implemented probably follows from that. Given that python has mature 1st class support from pytorch/jax frameworks ideally this project would find a space would be a space that's not well served by the standard frameworks (maybe prototyping apps/demos/visualizations/something else?) anyway feel free to join the discord if it's useful to brainstorm there.

Thank you for the feedback.
The first purpose is that we want to build a low-level wrapper for gemma.cpp to understand how it works.
The second one is for the local demo, apps on poor computational resource devices, we can utilize the speed of the framework and will make the wrapper's features good to develop around.
Can you send me the discord link for future brainstorm?

Link is here https://discord.gg/H5jCBAWxAe (it's now on the README)