Alexander Nair

Github Link

Made with: Python (Tensorflow), Rust

My One-Bit LLM project lets me dive into the deep-end of Artificial Intelligence. It's fun and also has real world application.

The aim is to develop a complete LLM from scratch. This includes working out all the maths myself. I'll also optimise it for space and speed so that it can run locally on any machine. To achieve this, I'll use a quantisation formula to convert each parameter into just one or two bits. This will reduce the size of the LLM and make the calculations lighter.

The eventual goal is to use quantisation formulas to convert other models to one-it. They can then be implemented onto embedded systems, which have memory and space restrictions (e.g. drones and smart machines).

This all matters because current LLMs use networking to communicate with these machines across the cloud. While this might work for a dishwasher, it wouldn't work in an airplane where latency matters. So for maximum safety, the LLM would need to exist onboard the airplane, hence the need for a 1-bit LLM.

Hi. I'm Alex (Alexander Nair).

I'm a university student studying for an MEng in Mathematics and Computer Science.

My interests are in the structure of Artificial Intelligence, real-world applications of AI, and research into AGI.

I like to create projects that mix mathematics and programming, basically trying to take code to new levels.

Along with my AI-related projects, I also create games, write stories, and act as the long-term Dungeon Master for a couple of DnD groups.