• minicpm-o

• openbmb/MiniCPM-o-2_6

A GPT-4o Level MLLM for Vision, Speech and Multimodal Live Streaming on Your Phone

GitHub | Online Demo | Technical Blog | Join Us

MiniCPM-o 2.6

This repository contains the MiniCPM-o 2.6 model weights in GGUF format, used for llama.cpp.

Currently, this readme only supports minicpm-omni's vision capabilities, and we will update the full-mode support as soon as possible.

Prepare models and code

Download MiniCPM-o-26 PyTorch model from huggingface to "MiniCPM-o-26" folder.

Clone llama.cpp:

git clone [email protected]:OpenBMB/llama.cpp.git
cd llama.cpp
git checkout minicpm-omni

Usage of MiniCPM-o 2.6

Convert PyTorch model to gguf files (You can also download the converted gguf by us)

python ./examples/llava/minicpmv-surgery.py -m ../MiniCPM-o-2_6
python ./examples/llava/minicpmv-convert-image-encoder-to-gguf.py -m ../MiniCPM-o-26 --minicpmv-projector ../MiniCPM-o-26/minicpmv.projector --output-dir ../MiniCPM-o-26/ --image-mean 0.5 0.5 0.5 --image-std 0.5 0.5 0.5 --minicpmvversion 4
python ./converthftogguf.py ../MiniCPM-o-26/model

quantize int4 version
./llama-quantize ../MiniCPM-o-26/model/ggml-model-f16.gguf ../MiniCPM-o-26/model/ggml-model-Q4KM.gguf Q4KM

Build llama.cpp using CMake:
https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md

cmake -B build
cmake --build build --config Release

Inference on Linux or Mac

# run f16 version
./llama-minicpmv-cli -m ../MiniCPM-o-26/model/ggml-model-f16.gguf --mmproj ../MiniCPM-o-26/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?"


run quantized int4 version
./llama-minicpmv-cli -m ../MiniCPM-o-26/model/ggml-model-Q4KM.gguf --mmproj ../MiniCPM-o-26/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"

or run in interactive mode
./llama-minicpmv-cli -m ../MiniCPM-o-26/model/ggml-model-Q4KM.gguf --mmproj ../MiniCPM-o-26/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -i