Llama 3.2 Vision

2024-11-05 · Source: Ollama Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, quick

Summary

Llama 3.2 Vision, a new multimodal large language model, was released on November 6, 2024, and is now available for local execution via Ollama. It comes in two sizes: an 11B parameter model requiring at least 8GB of VRAM, and a larger 90B parameter model needing a minimum of 64GB of VRAM. Users can download Ollama 0.4 and run the models directly from the command line. The model supports various vision-language tasks, including handwriting recognition, Optical Character Recognition (OCR), analysis of charts and tables, and general image question-answering. Integration is also provided through Ollama's Python and JavaScript libraries, as well as via cURL for API access.

Key takeaway

For Computer Vision Engineers developing local AI applications, Llama 3.2 Vision's availability in Ollama provides a robust, accessible option. You should consider its VRAM requirements (8GB for 11B, 64GB for 90B) when selecting a model size. This enables you to quickly prototype and deploy multimodal features like OCR or image Q&A without cloud dependencies, using familiar Python or JavaScript libraries.

Key insights

Llama 3.2 Vision offers local multimodal AI capabilities via Ollama, supporting diverse image-text tasks.

Principles

• Local execution of LLMs is feasible.
• VRAM capacity dictates model size usage.

Method

Download Ollama 0.4, then use `ollama run llama3.2-vision` (or `:90b`) to interact with the model, providing images via drag-and-drop or file path.

In practice

• Run Llama 3.2 Vision locally for image analysis.
• Integrate with Python/JavaScript for custom applications.

Topics

• Llama 3.2 Vision
• Multimodal AI
• Ollama
• Optical Character Recognition
• Image Question Answering

Code references

• user-attachments/assets
• ollama/ollama-python
• ollama/ollama-js

Best for: Computer Vision Engineer, Machine Learning Engineer, AI Engineer, Data Scientist

Related on AIssential

• Llama 3.2 goes small and multimodal — Llama 3.2 offers small, text-only models for edge devices and larger multimodal models for image reasoning.
• Llama 3 on Your Local Computer | Free GPT-4 Alternative — Llama 3 achieves significant performance gains through extensive training data, an improved tokenizer, and advanced fine-tuning.
• Llama 3 — Llama 3 offers substantial improvements over Llama 2, now locally runnable via Ollama.
• Introduction to Qwen3.5 – Overview, vLLM, and llama.cpp — Qwen3.5 supports image and video inference via vLLM and llama.cpp.
• What is your current local LLM setup? — The community actively explores diverse local LLM setups, balancing hardware, software, and model choices for varied applications.
• Gemma 4 12B MTP Local Test | Coding, OCR, Visual RAG with llama.cpp — The Gemma 4 12B MTP model offers multimodal capabilities and efficiency, nearing larger MoE models, but with performance trade-offs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Ollama Blog.