Ollama is based on the llama.cpp inference engine at its core and natively supports quantized models in GGUF format.