OpenBMB Releases MiniCPM-V-4.7-1B: An Efficient On-Device Multimodal Model
OpenBMB has released MiniCPM-V-4.7-1B, a highly efficient 1-billion-parameter multimodal language model optimized for on-device inference with enhanced single-image, multi-image, and video understanding capabilities alongside GGUF format support. This release advances the performance-efficiency Pareto frontier for small multimodal models, enabling powerful vision-language applications directly on resource-constrained edge devices and mobile hardware. MiniCPM-V 4.7 utilizes early visual token compression techniques and supports mixed 4x/16x compression, broad mobile and chip platform integration, and standard inference frameworks like llama.cpp and Ollama.
## BACKGROUND
MiniCPM-V is a series of pocket-sized multimodal large language models designed by OpenBMB for strong visual and textual understanding. GGUF is a binary file format optimized for fast model loading and efficient quantization, commonly used with llama.cpp to run large language models locally.