A performance-focused llama.cpp fork with new quantization formats and LLM inference/serving optimizations for local CPU and GPU deployment.