pingala8-llm

2-bit LLMs for llama.cpp: 85% smaller, beats IQ2_XXS at the same size, no retraining. One-click Windows build.

GitHub

Official source

Start locally with the CLI guide