Meta AI has announced the release of quantized Llama models, featuring increased speed and a reduced memory footprint. These models are designed for on-device and edge deployments, addressing the demand for more efficient and portable AI applications. The new models offer improved accuracy, faster inference, and reduced memory usage, making them suitable for mobile devices. This development is expected to enable unique experiences that prioritize privacy and reduce computational costs.

Source: https://ai.meta.com/blog/meta-llama-quantized-lightweight-models/?_fb_noscript=1