Llama 3.2 Vision available in Ollama

Llama 3.2 Vision is now available for local execution via Ollama, providing multimodal capabilities in 11B and 90B parameter sizes. This release enables users to process images and text simultaneously on their own hardware.

Model Sizes and Hardware Requirements

Llama 3.2 Vision is offered in two primary configurations to balance performance and resource availability:

  • 11B Model: Requires a minimum of 8GB of VRAM.
  • 90B Model: Requires a minimum of 64GB of VRAM.

Multimodal Capabilities

Llama 3.2 Vision supports a variety of visual understanding tasks. Key capabilities demonstrated include:

  • Handwriting Recognition: The model can interpret and transcribe handwritten text.
  • Optical Character Recognition (OCR): The model can extract printed text from images.
  • Charts and Tables: The model can analyze and interpret data presented in graphical formats or tabular structures.
  • Image Question and Answer (Q&A): The model can answer specific questions based on the visual content of a provided image.

Implementation and Usage

To use Llama 3.2 Vision, users must have Ollama 0.4 installed. The models can be pulled and executed via the command line using ollama run llama3.2-vision (for the 11B version) or ollama run llama3.2-vision:90b (for the 90B version).

Integration via Libraries

Ollama provides multiple ways to integrate Llama 3.2 Vision into applications:

  • Python: Using the ollama-python library, images can be passed within the messages array using the images key.
  • JavaScript: Using the ollama-js library, the chat method accepts an image path or base64 data within the message object.
  • cURL: The API allows for direct interaction via HTTP requests to http://localhost:11434/api/chat, where images are sent as base64-encoded data.

Local Interaction

In the terminal, users can add images to a prompt by dragging and dropping the image file or by specifying the file path to the image on Linux systems.

Sources

Related

  • Dispatch
    Llama 3.2 Support in Ollama

    Ollama now supports Meta's Llama 3.2, introducing lightweight 1B and 3B text-only models for edge devices and upcoming 11B and 90B vision-capable models.

  • Dispatch
    Llama 3.2 Release Notes: Multimodal Vision and On-Device Small Language Models

    Meta has released Llama 3.2, introducing multimodal vision capabilities in 11B and 90B sizes and lightweight 1B and 3B text-only models optimized for on-device deployment.

  • Dispatch
    Ollama LLaVA 1.6 Vision Model Support

    Ollama has integrated LLaVA 1.6, enhancing vision capabilities with higher image resolution, improved text recognition, and new model sizes including a 34B parameter version.

  • Dispatch
    Llama 3 Availability on Ollama

    Ollama has released support for Llama 3, providing access to the 8B and 70B parameter models with significant improvements in training data, context length, and token efficiency over Llama 2.

  • Dispatch
    Ollama Multimodal Engine Release

    Ollama has introduced a new multimodal engine to improve the reliability and accuracy of local inference for vision models, supporting Llama 4, Gemma 3, Qwen 2.5 VL, and Mistral Small 3.1.