Muse-Spark-Open 816B GGUF runs locally via llama.cpp on Mac hardware

Hugging Face CEO Clément Delangue posted a llama serve command that runs the ggml-org/Muse-Spark-Open-816B-A42B-GGUF model at Q2_K quantization on two Mac Studios, crediting @finkd and @johnternus for the setup.

Hugging Face's Merve Noyan later shared a llama.cpp command for the same Q2_K GGUF, saying it pairs well with the pi.dev agent and showing a demo running on a Mac M3 Ultra with 512GB of memory. Pedro Cuenca calls it a frontier model that runs on a desktop. The posts include commands and claims of running the model, but none reports throughput, latency, or output quality.