World's First LLM-powered Nintendo 64 Game — nano-GPT on-cart at 93MHz
Summary
This repo documents a nano-GPT language model running entirely on Nintendo 64 hardware (VR4300, 93.75 MHz) with 4 MB RAM. It uses Q8.7 fixed-point arithmetic and an SEAI binary format, demonstrating live LLM inference on vintage hardware and providing build, training, and integration details for libdragon-based projects. A notable case study in ultra-low-resource AI that may inspire edge AI explorations in constrained devices.