https://github.com/ggerganov/llama.cpp/issues/9246
thanks to :https://github.com/HimariO/llama.cpp/tree/qwen2-vl
Follow steps:
1 Download turingevo/Qwen2-VL-2B-Instruct-gguf
2 git clone https://github.com/HimariO/llama.cpp/tree/qwen2-vl
then build and get target llama-qwen2vl-cli
3 Get pictures:
It's recommended to resize the image to a resolution below 640x640, so it won't take forever to run on CPU backend:
ffmpeg -i input.jpeg -vf "scale=512:512" 1.png
4 cmd: llama-qwen2vl-cli -m Qwen2-VL-2B-Instruct-F16.gguf --mmproj qwen2vl-vision.gguf -p "Describe this image" --image "1.png"