15M Chat Model Implemented with Vanilla Commands
A fully functional chat model implemented with only commands, no mods or external API calls, everything's runned locally inside a huge datapack. It is transformer-based (specifically the LLaMA 2 variant of it), which is basically the same architecture that ChatGPT and other LLMs used. However in order to make it work with commands, I have to use a very small model (15 Million Parameters), and the speed is...well quite slow. Actually for this model it takes ~20 minutes to generate a single token, and I believe this demo took me around 24 hours to record (I basically took screenshots every 10 minutes and merged them together in the end). So yeah, just a cool demo, not useful for anything practical at all. You can also try this out for yourself if you're not bothered by the time. Here is the github repo of this project, you can find the download link in it and follow the instructions to use the datapack (I'd also be pretty happy if you give stars to it :D).
I've actually finished this project two months ago but forgot to put it here, but whatever :P