LLaMA Now Goes Faster on CPUs