Technical report: Read paper · PDF · BibTeX · All models
Collection: Mermaid models
Usage: Step-by-step Windows, Linux and macOS instructions.
MiniCPM-V 4.7 1B Mermaid Heretic — GGUF Q4_K_M
Heretic-modified MiniCPM-V 4.7 1B Mermaid GGUF artifacts for local vision-language inference. The repository contains a Q4_K_M language model and an F16 vision projector; load both for image input.
| File | Size (decimal GB) |
|---|---|
mmproj-model-f16.gguf |
1.109 |
model-Q4_K_M.gguf |
0.672 |
Download and run
pip install -U huggingface_hub
hf download Lucasgitguy-co/minicpm-mermaid-heretic-gguf --local-dir ./minicpm-mermaid-heretic-gguf
Use a llama.cpp build that supports MiniCPM-V 4.7 and its projector architecture. With the downloaded files:
llama-mtmd-cli \
-m ./minicpm-mermaid-heretic-gguf/model-Q4_K_M.gguf \
--mmproj ./minicpm-mermaid-heretic-gguf/mmproj-model-f16.gguf \
--image diagram.png \
-p "Convert this diagram into Mermaid code."
The command is an example and was not inference-tested as part of this documentation update. Check your runtime’s MiniCPM-V 4.7 support before downloading.
One-file bundle
minicpm-mermaid-heretic-bundle.gguf (1.781 GB) holds model-Q4_K_M.gguf and mmproj-model-f16.gguf in one file, byte for byte. The name has no quant tag or "mmproj" in it, so llama.cpp -hf, Ollama and LM Studio keep picking the separate files. Use it if you'd rather download one file. llama.cpp can't run the bundle directly, because it always needs the model and mmproj as two files. Unpack it with gguf_bundle.py (Python 3, no packages needed):
hf download Lucasgitguy-co/minicpm-mermaid-heretic-gguf minicpm-mermaid-heretic-bundle.gguf gguf_bundle.py --local-dir .
python3 gguf_bundle.py unpack minicpm-mermaid-heretic-bundle.gguf -o .
The script checks each file's sha256 while it unpacks. python3 gguf_bundle.py list <bundle> shows what's inside.
Checked 2026-11-16: the unpacked files have the same sha256 as the separate files in this repo.
Related checkpoint
Evaluation and provenance
This card describes the repository files and configuration. Detailed training data, training procedure, and benchmark results are not documented in the inspected artifacts. Loading examples are guidance, not a claim of measured model quality.
License and attribution
This release is licensed under Apache 2.0; see LICENSE and NOTICE. Upstream attribution and applicable third-party notices must be retained when redistributing. The license of the Heretic tool is separate from the license of the model artifacts.
Repository guide
Related project data: text and image collection and original English/Dutch text collection. These links describe related project resources; they do not establish the exact training split of every checkpoint.