Breaking large files into smaller related files to grow the contributor community

[jmagder]

Good afternoon (CST), and happy Monday! I want to start by saying what an incredible community we have here. I've developed refresh OCD to see the exciting changes made throughout the day. I've learned things by looking at PRs, and started poking around the code myself.

As a local llama.cpp user, I would love to be able to analyze the code, and eventually make my own contributions. LLMs are a great tool for helping with this, but their limited context window (especially in a local environment) means that we have files that no LLM can actually read into its context. This made me think, perhaps we could make the process of learning and improving the code base more tractable by splitting related functions into their own files? Then we could load only the necessary parts into context and have a more productive time understanding and improving the codebase.

The hardest part would be coordinating with some other very large PRs. I know @pwilkin has done some amazing work, and getting that merged first seems more important. I could open a PR with some changes, but I'd love to hear from the community on the best way to orchestrate this so we don't step on each others toes. Thanks for your input everyone, lets keep building this incredible project! 🥳

[pwilkin]

I'll revisit the split once I'm done with Qwen3 Next. @ngxson has provided some feedback, so I'll first rework the PR to accomodate his feedback and hopefully we can move from there.

FWIW the separate PR for the llama-model.cpp split is here:
#16252

[jmagder]

Thank you for sharing that PR! You are far further along with this than I realized. Thank you for all you do! Let me know how I can help.