Llama 3.2

Released by Meta in 2024, the Llama 3.2 family introduces multimodal vision capabilities and lightweight 1B and 3B parameter variants. This allows the models to process images and run efficiently on edge devices.

Yes, the 1B and 3B parameter models in the Llama 3.2 family are specifically optimized for on-device and edge deployment. Meta partnered with hardware manufacturers like Qualcomm and MediaTek to ensure the models run smoothly on mobile phones.

Yes, like its predecessors, Meta released Llama 3.2 as an open-weight model family. This allows developers and researchers to download the weights for free, though there are specific acceptable use policies in place.

The larger Llama 3.2 models (11B and 90B) are multimodal and use a vision encoder to process images alongside text. This allows developers to build applications that can answer questions about photos or analyze visual data.