Sub-15ms Local Decisions: Running Laya Models with Native MLX on Apple Silicon

TL;DR Stop burning tokens and introducing multi-second cloud latency just to make...