I’ve been working on NanoGPTForge, a modified version of Andrej Karpathy's nanoGPT that emphasizes simplicity, clean code, and type safety, while building directly on PyTorch primitives. It’s designed to be plug-and-play, so you can start experimenting quickly with minimal setup and focus on training or testing models right away.

Contributions of any kind are welcome, whether it is refactoring code, adding new features, or expanding examples. I’d be glad to connect with others interested in collaborating!

Check it out here: https://github.com/SergiuDeveloper/NanoGPTForge