---
base_model:
- openbmb/MiniCPM-V-4.7-1B
library_name: transformers
tags:
- text-generation-inference
- llama-cpp
- minicpm-v
- multimodal
- On-Device Model
- lightweight
license: apache-2.0
language:
- en
pipeline_tag: image-text-to-text
---

# **openbmb-MiniCPM-V-4.7-1B-GGUF**

> **MiniCPM-V 4.7 1B** is the latest model in OpenBMB's MiniCPM-V series, a roughly 1B-parameter multimodal model for single-image, multi-image and video understanding, designed for on-device use. It improves on MiniCPM-V 4.6's foundation capabilities and, in Instruct mode, is said to substantially outperform Qwen3.5 0.8B in both overall score and inference efficiency, which the card presents as advancing the performance-efficiency Pareto frontier. It uses intra-ViT early compression from LLaVA-UHD v4, which compresses visual tokens inside the vision encoder to cut computation in later layers. It also supports mixed 4x/16x visual token compression, so developers can trade visual detail against cost. The benchmark comparisons appear only as images, so the individual scores aren't readable from the text. The acknowledgements credit Qwen3.6 as the language backbone and SigLIP2 as the vision module. Deployment targets include iOS, Android and HarmonyOS, with day-0 support for Rockchip, Houmo, Intel, AMD and Arm chips and an OpenVINO path for Intel hardware. It loads through Transformers (>=5.18.0), can be served with `transformers serve`, and is supported by SGLang, vLLM, llama.cpp and Ollama. Fine-tuning works with SWIFT and LLaMA-Factory, and the card introduces Self-Replay Fine-Tuning, meant to extend capabilities cheaply while reducing forgetting. Variants include int4 (bitsandbytes), AWQ, GPTQ, GGUF and FP8 quantizations, plus a DSpark draft model for speculative decoding. It is released under Apache 2.0.

## Model Files

| File Name | Quant Type | File Size | File Link | Description |
|-----------|------------|-----------|-----------|-------------|
| MiniCPM-V-4.7-1B.BF16.gguf | BF16 | 2.02 GB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.BF16.gguf) | Full BF16 weights. Highest quality, largest file size. |
| MiniCPM-V-4.7-1B.MXFP4_MOE.gguf | MXFP4_MOE | 1.08 GB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.MXFP4_MOE.gguf) | Mixed precision / MoE optimized quantization, very small size. |
| MiniCPM-V-4.7-1B.Q3_K_L.gguf | Q3_K_L | 600 MB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.Q3_K_L.gguf) | Lower quality but usable, good for low RAM availability. |
| MiniCPM-V-4.7-1B.Q3_K_M.gguf | Q3_K_M | 575 MB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.Q3_K_M.gguf) | Low quality. |
| MiniCPM-V-4.7-1B.Q4_K_M.gguf | Q4_K_M | 672 MB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.Q4_K_M.gguf) | Good quality, default size for most use cases, *recommended*. |
| MiniCPM-V-4.7-1B.Q4_K_S.gguf | Q4_K_S | 648 MB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.Q4_K_S.gguf) | Slightly lower quality with more space savings, *recommended*. |
| MiniCPM-V-4.7-1B.Q5_K_M.gguf | Q5_K_M | 752 MB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.Q5_K_M.gguf) | High quality, *recommended*. |
| MiniCPM-V-4.7-1B.Q5_K_S.gguf | Q5_K_S | 738 MB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.Q5_K_S.gguf) | High quality, *recommended*. |
| MiniCPM-V-4.7-1B.Q6_K.gguf | Q6_K | 838 MB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.Q6_K.gguf) | Very high quality, near perfect, *recommended*. |
| MiniCPM-V-4.7-1B.Q8_0.gguf | Q8_0 | 1.08 GB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.Q8_0.gguf) | Extremely high quality, generally unneeded but max available quant. |
| MiniCPM-V-4.7-1B.mmproj-bf16.gguf | mmproj-bf16 | 1.11 GB | [Link](https://huggingface.co/prithivMLmods/openbmb-MiniCPM-V-4.7-1B-GGUF/blob/main/MiniCPM-V-4.7-1B.mmproj-bf16.gguf) | Multimodal projection file in BF16 format. Used for vision/language models. |

## llama.cpp

LLM inference in C/C++ — https://github.com/ggml-org/llama.cpp