CPU-first. GPU-friendly. Terllama brings ternary quantization to llama.cpp—run LLMs anywhere. - View it on GitHub