Meta has introduced two new quantized models, Llama 3.2 1B and 3B, designed for mobile devices. These innovative models achieve 2-4 times faster inference speeds, with a 56% reduction in model size and 41% less memory usage compared to prior iterations. Utilizing Quantization-Aware Training combined with LoRA adaptors, they cater to the rising need for efficient AI on resource-limited devices. Available on Meta's platform and Hugging Face, these models aim to broaden access to powerful AI development for developers everywhere.