ocrcpm is a pip-installable document-parsing SDK for MiniCPM-V OCR. It wraps the whole pipeline — layout detection, crop preparation, per-crop recognition and Markdown assembly — behind a single command line, and keeps the inference engine pluggable so the same pipeline runs on Transformers, vLLM, SGLang, llama.cpp or Ollama.
| Model | English | 中文 |
|---|---|---|
| MiniCPM-V OCR | Guide | 指南 |
| Path | Description |
|---|---|
ocrcpm-0.1.0-py3-none-any.whl |
Prebuilt wheel, install with pip install "./ocrcpm-0.1.0-py3-none-any.whl[all]" |
wheel_source/ |
Source behind that wheel — pyproject.toml, MANIFEST.in and src/ocrcpm/; rebuild with python -m build |
The OCR model weights are not part of the wheel and are not open-sourced yet; the guides use placeholder model paths for now.