llama.cpp Release b11551 Adds Support for MiniCPM-V 4.7
llama.cpp release b11551 adds GGUF conversion and multimodal inference support for the MiniCPM-V 4.7 model. The release includes updates to Python model conversion scripts, adjustments for positional embeddings, and new GGUF metadata attributes. This update enables developers and researchers to run MiniCPM-V 4.7 locally across a wide range of hardware, including mobile and edge devices. It reflects the ongoing trend of bringing lightweight vision-language capabilities directly to consumer-grade hardware without relying on cloud APIs. Technical additions include handling grid layouts, supporting Multimodal Rotary Position Embedding (mRoPE) time axes from extra position slots, and adding `rope.section_order` metadata to GGUF format.
## BACKGROUND
llama.cpp is a popular open-source C/C++ inference framework designed to run large language models and multimodal vision-language models locally using quantized GGUF format. MiniCPM-V is an efficient series of vision-language models developed by OpenBMB that supports image and text understanding for edge deployments.