Idea: Mathematical Enhancements for GPT/LLM Architectures (teaser)

[muzamil-ashiq]

Hi llama.cpp community 👋,

I’ve been prototyping a conceptual extension to GPT/LLM-style models—using mathematical mechanisms aimed at improving memory handling and adaptive learning—while keeping the core structure lightweight.

The work is written up in Markdown (README + comparison), but I haven’t published the files yet. Before doing so, I’d love to know:

• Could this type of idea be relevant or interesting for llama.cpp experiments?
• Would you be open to taking a look privately before I share more widely?

Appreciate your feedback 🚀