Meta releases compact versions of the Llama 3.2 AI models

The new quantized models are 56 percent smaller and use 41 percent less memory compared to the full-size models released last month.

Meta has released compact versions of its lightweight Llama 3.2 1B and 3B models that are small enough to run effectively on mobile devices.

The Facebook owner said in an announcement yesterday (October 24) that these new “quantized” models are 56 percent smaller and use 41 percent less memory compared to the original 3.2 models released last month.

Meta says you can use the 1B or 3B models for on-device applications such as: B. to summarize a discussion from your phone or to access tools on the device like the calendar.

The new models “apply the same quality and security requirements” as the original Llama 3.2 1B and 3B, but process information two to three times faster, the company said.

The new versions were quantified using two techniques – one that prioritizes accuracy in low-precision environments and one that prioritizes portability while aiming to maintain quality.

“These models offer lower memory footprints, faster on-device inference, accuracy and portability – all while maintaining quality and security for developers when deploying to resource-constrained devices,” the announcement said.

Users will be able to download and deploy these new model versions on mobile CPUs that the company developed in “close collaboration” with other industry leaders in this space, it said.

These lightweight models are part of the text-only series of 1B and 3B models available in the EU.

The multimodal models 11B and 90B, which can handle multiple formats such as text, images, audio and video, are not available in the EU. In the summer, Meta said it would not launch these models in the EU due to the bloc’s “unpredictable” regulatory environment.

A month before this announcement, after intensive discussions with the Irish Data Protection Commission, the company withdrew its plans to train its large language models on public content shared by adults on Facebook and Instagram.

Data protection group Noyb raised serious concerns about the plan, claiming that Meta’s intention to use AI training material that comes from public and licensed data and could contain personal information would violate the GDPR.

Don’t miss out on the knowledge you need to succeed. Sign up for Daily letterSilicon Republic’s roundup of top science and technology news.