I've been tinkering with how small language models can be while still doing something interesting. This led me to create MiniLMs: a website and open-source repository where I'm experimenting with minimalist LLM architectures and sharing my learning process.

The current main focus is "SYNEVA," a small Markov chain chatbot I worked on – some with pattern matching to simple neural nets, always aiming for a tiny <3kb footprint. You can try it (and some older versions) on the site.

The project also includes: Study resources I'm gathering on NNs, LLMs, and AI history. the current models are HTML/JS. Definitely a learning-in-public kind of thing.

Website & Demos: https://minilms.kuber.studio/ GitHub Repository (MIT License): https://github.com/Kuberwastaken/MiniLMs-PROJECT

Curious to hear your thoughts or any cool tiny AI projects you've seen or want to see.