I made a program that converts any llama2 large language model into a minecraft datapack, and you can run inference right inside the game. It's still semi-finished, Currently I've only implemented argmax sampling, so the output tends to stuck in loops sometimes. Adding top-p sampling will probably i
