Manchu LLaMA 3.2 11B Vision Model

This repository contains scripts to run the dhchoi/manchu-llama32-11b-vision-merged model for Manchu script OCR and text generation.

Note: This application is optimized for macOS with Apple Silicon (MPS) or CPU support only.

Setup

1. Clone Repository

git clone https://github.com/dhchoi-lazy/manchu_mac.git
cd manchu_mac

2. Clean Virtual Environment (if exists)

# Remove existing virtual environment for clean setup
rm -rf .venv

3. Create Virtual Environment

python -m venv .venv
source .venv/bin/activate

4. Install Dependencies

pip install -r requirements.txt

Usage

Web Interface (Streamlit)

The easiest way to use the Manchu OCR model is through the web interface:

python run_streamlit.py

The web app will open automatically in your browser at http://localhost:10011

Command Line Interface

Manchu OCR Model

# Interactive mode
python run_manchu_model.py

# Single image OCR
python run_manchu_model.py ./samples/validation_sample_0000.jpg

# Batch directory processing
python run_manchu_model.py ./samples/

Model Evaluation

python run_manchu_eval.py

Model Information

• Model: dhchoi/manchu-llama32-11b-vision-merged
• Base: LLaMA 3.2 11B Vision
• Purpose: Manchu script OCR and text generation
• Size: ~21.3 GB
• Device Support: Apple Silicon (MPS) or CPU

System Requirements

• OS: macOS (Apple Silicon recommended)
• Python: 3.10+
• Memory: 16GB+ RAM recommended
• Storage: 25GB+ free space for model cache

Performance

• Apple Silicon (MPS): ~30-40 seconds per image
• CPU: ~60-120 seconds per image (depending on CPU)