Metas Llama 3.2 AI Can Run Natively and Understand Images

Meta has launched the next generation of free and open source master’s courses. The new Llama 3.2 models can be run natively (even on mobile devices) and now have image processing capabilities. Vision-supported multimodal LLMs that are also open source and free are still rare.

Llama 3.2 comes in two mid-weight and two lightweight models. 90B and 11B Master’s It can process images such as photos, documents, charts and graphs. “For example, a person might ask a question about which month of the previous year their small business had the best sales, and then Llama 3.2 could reason based on an existing graph and provide the answer quickly. In another example, the model could reason with a map and follow the walk.” It can help answer questions like when it might be steeper or the distance of a particular trail marked on the map,” Meta explained of the new feature.

You can also ask these models to describe an image in detail or create stories or titles based on the description. The performance of these two models is comparable. ChatGPT 4o Mini and Claude 3 Haiku.

The lighter 1D and 3D Llama 3.2 models can run natively on several edge devices (such as smartphones and PCs). Meta has partnered with Qualcomm to run Llama 3.2 models natively on Snapdragon 8 Gen 3 and Snapdragon X series chips. You can use these platforms on premium Android phones and new generation Copilot+ PCsrespectively. These two models do not have image recognition, but they can handle up to 128 thousand tokens, allowing them to process long documents and sustain conversations for longer periods of time without losing context.

Running these models on the device is better in terms of privacy and performance. Your data is never sent to the cloud, so an app using this model is more private by default. Since the process does not involve the internet, latency is minimal. Meta says responses on the device “feel immediate.”

Llama 3.2 models can be downloaded from: lama.com And Hugging FaceYou can try them out on a desktop computer via LM Studio (just search for “Llama-3.2” in the Explore tab).

Source: Meta