MiniCPM-V-4.6 is a compact vision-language model developed by OpenBMB, designed for efficient multimodal understanding across images, documents, charts, diagrams, screenshots, and natural scenes. This repository contains GGUF quantized variants of the model optimized for efficient local inference using llama.cpp.

Unlike task-specific OCR or document parsing models, MiniCPM-V-4.6 is a general-purpose multimodal assistant capable of jointly understanding visual and textual information. It combines image perception, reasoning, OCR capability, document understanding, and multimodal dialogue within a compact architecture suitable for resource-efficient deployment.

The quantized formats significantly reduce memory requirements while preserving strong multimodal reasoning capability, enabling practical deployment on consumer hardware and edge AI systems.

Model Overview

Model Name: MiniCPM-V-4.6
Base Model: openbmb/MiniCPM-V-4.6
Architecture: Vision-Language Model (VLM)
Modalities: Text, Image
Primary Languages: Multilingual
Developer: OpenBMB
License: Apache 2.0

Quantization Formats

This repository provides various GGUF quantized versions of the MiniCPM-V-4.6 model optimized for efficient local inference using llama.cpp.

IQ3_M

Size reduction of approx 70.05% (432.74 MB) compared to 16-bit (1.41 GB)
Aggressive 3-bit quantization optimized for lightweight multimodal inference on memory-constrained hardware
Suitable for image understanding, visual question answering, OCR, and multimodal conversational workloads
Enables efficient deployment on consumer CPUs and lower-memory GPU systems
Fine-grained visual reasoning and complex scene interpretation may experience reduced fidelity compared to higher-precision variants

IQ4_NL

Size reduction of approx 65.84% (493.22 MB) compared to 16-bit (1.41 GB)
Advanced 4-bit non-linear quantization designed to better preserve multimodal reasoning and visual understanding quality
Better suited for image analysis, document interpretation, OCR, and chart understanding workflows
Provides improved consistency across diverse multimodal tasks while minimizing quantization loss
May require slightly increased computational overhead during inference

IQ4_XS

Size reduction of approx 66.61% (482.19 MB) compared to 16-bit (1.41 GB)
Balanced 4-bit quantization optimized for efficient multimodal inference and dependable response quality
Provides a practical balance between memory efficiency, visual understanding performance, and runtime speed
Suitable for multimodal assistants, image reasoning, document understanding, and edge AI deployments
Maintains stable performance across a broad range of real-world vision-language applications

Training Background (Original Model)

MiniCPM-V-4.6 is trained with an emphasis on multimodal reasoning, image understanding, OCR, document comprehension, and visual-language alignment across diverse image and text datasets.

Pretraining

Large-scale multimodal pretraining using image-text datasets spanning natural images, documents, charts, screenshots, and diagrams
Focus on visual-language alignment, semantic understanding, and multimodal representation learning
Optimized for downstream multimodal reasoning and visual understanding tasks

Multimodal Optimization

Enhanced for visual reasoning, document understanding, OCR, chart interpretation, and multimodal dialogue
Improved performance across image analysis, visual question answering, and structured multimodal interactions
Designed to provide strong multimodal capability while maintaining an efficient model footprint

Key Capabilities

Image Understanding Interprets natural images, scenes, screenshots, and visual content.
Visual Reasoning Performs reasoning over complex visual information and image-based tasks.
Document Understanding Understands documents, reports, tables, and structured visual layouts.
Optical Character Recognition (OCR) Extracts and interprets textual information embedded within images.
Chart & Diagram Analysis Understands charts, plots, diagrams, and other structured visual representations.
Multimodal Conversation Supports interactive conversations combining image and text inputs.
Efficient Local Deployment Quantized variants enable practical multimodal inference on consumer hardware.

Usage Example

Using llama.cpp

bash

./llama-mtmd-cli \
  -m SandLogicTechnologies/MiniCPM-V-4.6_IQ4_NL.gguf \
  --mmproj SandLogicTechnologies/mmproj-minicpm-v-4.6-f16.gguf \
  --image chart.png \
  -p "Analyze this chart and explain the key trends."

Recommended Usecases

Vision-Language Assistants Build interactive multimodal AI assistants capable of understanding images and text.
Visual Question Answering Answer questions based on photographs, diagrams, charts, and screenshots.
Document & OCR Applications Interpret documents and extract textual information from visual content.
Chart & Diagram Interpretation Analyze graphs, plots, technical diagrams, and presentation materials.
Educational & Research Tools Support multimodal learning, tutoring, and visual content understanding.
Edge AI Deployments Deploy compact multimodal intelligence on resource-constrained hardware.

Acknowledgments

These quantized models are based on the original work by the *OpenBMB- development team.

Special thanks to:

The OpenBMB team for developing and releasing the MiniCPM-V-4.6 model.
Georgi Gerganov and the llama.cpp open-source community for enabling efficient quantization and inference through the GGUF format.

Contact

For questions, feedback, or support, please reach out at support@sandlogic.com or visit https://www.sandlogic.com/

SandLogicTechnologies/MiniCPM-V-4.6-GGUF

MiniCPM-V-4.6