BitNet on CPUs: Conversational Bots Without GPUs How 1.58‑bit LLMs make private, affordable, and fast-enough AI practical on commodity servers