Meta has introduced quantized versions of its Llama 3.2 models, enhancing on-device AI performance with up to four times faster inference.