ggml-org/llama.cpp

Known for

Running quantised models on hardware you already own, including a laptop CPU.

The inference engine almost everything local is ultimately built on, including Ollama. Its own star velocity understates how much of the ecosystem depends on it.