This repository provides instructions and scripts to run Bonsai quantized language models locally on Mac (Metal), Linux/Windows (CUDA, Vulkan, ROCm), or CPU. It includes pre-built llama.cpp binaries and MLX (Apple Silicon) forks with support for Q1_0 and Q1_58 quantization formats. Model weights are distributed via HuggingFace collections, with companion web demos and Google Colab notebooks for easy experimentation.