Apple ML Research: Muse Spark Open with MLX on Mac Studio
Summary
Apple ML Research showcases local and distributed LLM inference on Mac Studio using MLX and RDMA, enabling frontier-class models like Muse Spark Open to run privately on Apple silicon. The article profiles performance across single and multi-node setups, various context sizes, and parallelization schemes (pipeline and tensor), with metrics on TTFT, throughput, and power consumption, highlighting the integration of MLX with Thunderbolt RDMA and Apple’s hardware accelerators.