Untukmu.AI

Meta's Llama model uses split inference: 1st layer processes on user devices, remaining 31 in cloud, enabling privacy while avoiding quantisation