MLX

Array framework for machine learning on Apple Silicon. NumPy-like API with lazy computation, dynamic graph construction, and unified memory (arrays live in shared CPU/GPU memory). C++, Python, and Swift bindings.

25K+ GitHub stars. 3x faster than llama.cpp on Apple hardware for some workloads. Enables efficient local LLM inference on MacBooks and has become the standard framework for running open models on Apple hardware. MIT License.