Meta Platforms releases Llama 3.2 lightweight AI model suitable for mobile devices
Meta unveiled the Llama 3.2 large-scale language model at the Meta Connect 2024 conference, including lightweight models with 1B and 3B parameters for mobile devices, supporting text input and a context length of 128K. The model is optimized for Qualcomm and MediaTek hardware, aiming to help developers build personalized device assistant applications, ensuring privacy and fast response. Llama 3.2 also includes 11B and 90B models, supporting text and image input, with visual understanding capabilities
According to the financial news app Zhitong Finance, the social media giant Meta (META.US) unveiled its latest large-scale language model Llama 3.2 at the Meta Connect 2024 conference on Wednesday. The model includes visual language models with 90B and 11B parameters, as well as lightweight pure text models with 1B and 3B parameters, the latter of which is suitable for edge and mobile devices.
It is reported that the 1B and 3B models of Llama 3.2 only support text input and have a context length of 128K. These models can perform use cases on devices such as summarization, instruction tracking, and local rewriting tasks.
These models support Qualcomm (QCOM.US) and MediaTek hardware, and are optimized for Arm (ARM.US) processors.
The 11B and 90B models of Llama 3.2 support both text and image inputs, with image and visual understanding capabilities.
The Llama 3.2 model aims to be the most developer-centric model to date, helping developers build device agents.
Meta stated: "These models enable developers to build personalized device agent applications that are highly privacy-preserving, with data never leaving the device." "For example, such applications can help summarize the last 10 messages received, extract action items, and use tools to call and directly send calendar invites for follow-up meetings."
As the processing is done locally, models on devices can achieve faster response times and help ensure privacy, as information does not need to be migrated to the cloud.
Meta is not the first company to introduce device-embedded artificial intelligence models. Google (GOOG.US, GOOGL.US) latest Pixel 9 Pro smartphone features Gemini AI. Apple (AAPL.US) iPhone 16 comes with Apple Intelligence and will provide device-embedded AI after the release of iOS 18.1 next month.
In addition, Meta has also launched the first pair of augmented reality (AR) glasses Orion and the latest virtual reality (VR) head-mounted device Quest 3S.
The above news drove Meta's stock price up over 2% to $575.77 at one point, setting a new all-time high