Ollama 0.4 brings Llama 3.2 Vision locally with one command
In one sentence With version 0.4 Ollama supports Meta's Llama 3.2 Vision: 'ollama run llama3.2-vision' runs the 11B model locally (at least 8 GB of VRAM), the 90B variant needs at least 64. Images go in the prompt or in the API images field.
Ollama is the program that makes it simple to run AI models on your own computer: one command and the model is ready. Since February 2024 it already supported some models able to "see" images, such as LLaVA.
On November 6, 2024, with version 0.4, Meta's multimodal Llama 3.2 Vision arrives. Just type ollama run llama3.2-vision and then drag an image into the terminal (or give its path) to ask the model to describe it, read a document or analyze a chart.
Everything runs locally, with no API keys and without sending images to a cloud service. The 11-billion-parameter model needs at least 8 GB of video memory; the 90-billion version at least 64 GB.
Companies
Ollama, Meta
Tools
Ollama, Llama 3.2 Vision
Tags
Sources