This project ports MiniGPT-4, a vision-language model, into pure C/C++ using the GGML tensor library for efficient inference. It supports multiple quantization levels (4-bit through 16-bit) to enable running large models on CPU with reduced memory footprint. The implementation includes a web UI and can load and preprocess images within the library.