llava-cpp-offline 🧠🖼️

Offline image analysis with LLaVA 1.6 + Visual Studio build
by Baris

This project shows how to compile and run LLaVA locally on Windows – without any API, without cloud, and without Python dependencies.

🔥 Features

• Fully offline LLaVA Vision model
• Compiled llama-llava-cli (.exe)
• GGUF model support
• GUI + image understanding
• Python wrapper via llava_cpp.py

📦 Repository contents

• llama-llava-cli.exe → compiled CLI interface for LLaVA
• llava_cpp.py → Python wrapper for CLI calls
• example_usage.py → test script
• models/ → place your LLaVA + MMProj models here
• screenshots/ → test images

🧠 Download models

👉 Main model:
https://huggingface.co/liuhaotian/llava-v1.6-mistral-7b-GGUF

👉 MMProj encoder:
https://huggingface.co/liuhaotian/llava-v1.6-mistral-7b-GGUF/tree/main/mmproj

Put both files inside the models/ folder:

🐍 Python example

from llava_cpp import analyze_image

response = analyze_image("screenshots/test_gui.png")
print("Response:", response)