Building an LLM replacement

GitHub repo here: github.com/jrz97619761/test-model-thing LLMs have a lot of problems, so I designed and built a small proof-of-concept language model architecture (not an LLM) that incorporates the following: Latent-space prediction Internal state + recurrent trace units (RTUs) Byte input and output Continuous data streaming Test-time / continual training Feel free to comment here, open issues, and fork the repo to improve my architecture or ask any questions! Share the video with people you think might be interested! Thanks :)