---

license: apache-2.0

base_model: openbmb/MiniCPM-V-4.7-1B

base_model_relation: quantized

library_name: llama.cpp

pipeline_tag: image-text-to-text

tags:

• gguf
• llama.cpp
• minicpm-v
• quantized
• multimodal
• image-text-to-text

---

MiniCPM-V 4.7 1B GGUF

Community GGUF quantizations of openbmb/MiniCPM-V-4.7-1B.

<div align="center" style="background-color:#f59e0b;color:#ffffff;padding:16px 20px;border-radius:10px;line-height:1.7;">

☕ If this GGUF made your day easier, a coffee would make mine.<br>

<a href="https://ko-fi.com/ngquocvinh" style="color:#ffffff;"><strong style="color:#ffffff;">Send a coffee ☕</strong></a><br>

I build and test these releases myself. Your coffee helps keep me going.<br>

Thank you for supporting this work.

</div>

About MiniCPM-V 4.7

MiniCPM-V 4.7 1B is OpenBMB's compact multimodal model for image and video understanding. The upstream model describes single-image, multi-image, and video use, with an efficient vision encoder and on-device deployment targets. See the official model card and the upstream deployment guide.

The upstream configuration declares a 262,144-token text context. The example below uses a 4,096-token context; long-context behavior has not been evaluated in this release. This GGUF release keeps the vision projector in F16 and quantizes the language-model weights. The default upstream 16x visual-token compression is used.

<a href="https://huggingface.co/openbmb/MiniCPM-V-4.7-1B"><img src="https://raw.githubusercontent.com/OpenBMB/MiniCPM-o/main/assets/minicpmv4.7/performance_efficiency_frontier.png" alt="Performance versus inference time for MiniCPM-V 4.7 1B"></a>

Quick start

With a CPU build of llama.cpp, ask a question about a local image:

./llama-mtmd-cli \
  -m MiniCPM-V-4_7-1B-Q4_K_M.gguf \
  --mmproj mmproj-model-f16.gguf \
  --chat-template-file chat_template.jinja \
  --image ./photo.jpg \
  --jinja \
  --simple-io --single-turn \
  -p 'Describe the image in one sentence.' \
  -n 128 -c 4096

Reproducibility and validation

The locked upstream BF16 checkpoint is converted to one F16 GGUF, and every quantized language-model GGUF is quantized directly from that same F16 input using a MiniCPM-V 4.7 1B-specific imatrix. No quantized GGUF is used as the source for another quantization. The projector stays in F16. The reproducibility manifest records source and runtime revisions, hashes, commands, and validation details. Artifact measurements use decimal GB (bytes / 1e9); SHA256SUMS.txt covers the published files.

Each published quantization is loaded with llama.cpp and must generate non-empty output for text and a single-image prompt. This is a compatibility smoke test, not a quality benchmark. No hold-out fidelity evaluation is published, so this card does not rank quantizations or make quality recommendations. The 2-bit, IQ, and ternary variants are experimental builds without a quality claim.

License and attribution

The upstream model is licensed under Apache 2.0. These are community GGUF quantizations, not an official OpenBMB release or endorsement.