Meta(META.US) releases Llama 3.2 lightweight AI model for mobile devices

Meta(META.US) released its latest large language model Llama3.2 at Meta Connect 2024 on Wednesday, including a 90B and 11B parameter visual language model and a lightweight text-only model with 1B and 3B parameters, which is suitable for edge and mobile devices.

It is known that the Llama3.21B and 3B models only support text input and support a context length of up to 128K. These models can perform use cases on devices such as summarization, command tracking, and local rewriting tasks.

These models support Qualcomm(QCOM.US) and MediaTek hardware and are optimized for Arm(ARM.US) processors.

Llama3.211B and 90B models support text and image inputs and have image and visual understanding capabilities.

Llama3.2 models are designed to be the most developer-centric models to date and help developers build device agents.

Meta said: "These models enable developers to build personalized device agent applications with strong privacy, with data never leaving the device." "For example, such applications can help summarize the last 10 messages received, extract operations, and leverage tools to call a calendar invitation directly for a follow-up meeting."

Since the processing is done locally, the models on devices can achieve faster response and help ensure privacy, as information does not need to be migrated to the cloud.

Meta is not the first company to launch AI models built into devices. Google's (GOOG.US, GOOGL.US) latest Pixel 9 Pro smartphone is equipped with Gemini AI. Apple's (AAPL.US) iPhone 16 is equipped with Apple Intelligence, which will be available with device-built AI after the release of iOS 18.1 next month.

Moreover, Meta also launched its first augmented reality (AR) glasses Orion and the latest virtual reality (VR) headset device Quest 3S.

The above news pushed Meta's stock price up more than 2% to a record high of $575.77.