AI & Compute roundup: Hugging Face runs Meta’s 816B open model at 18 tokens a second on one 512GB Mac
Hugging Face ran Meta's 816B Muse Spark Open at 2-bit on one 512GB M3 Ultra Mac. Reflection's 501B Beam needs about 280 GB at 4-bit, and NVIDIA's Bolt 74B-A9B is built to fit 24GB.