---
license: apache-2.0
base_model: PaddlePaddle/PaddleOCR-VL-1.6
tags:
  - gguf
  - ocr
  - handwriting
  - llama.cpp
---

# PenHour handwriting reader

The handwriting reader used by [PenHour](https://penhour.com), a Mac app for journaling on paper.
PenHour downloads these two files once and runs them on the user's Mac with
[llama.cpp](https://github.com/ggml-org/llama.cpp). Pages are never uploaded.

| File | What it is | Size | SHA-256 |
|---|---|---|---|
| `PaddleOCR-VL-1.6-Q5_K_M.gguf` | Language model, Q5_K_M | 341 MB | `5d775957de8da2fded5b488b78743f6a48203377a391913d909372d7ef957ad6` |
| `PaddleOCR-VL-1.6-mmproj-Q4_K_M.gguf` | Vision projector, Q4_K_M | 487 MB | `e2bd793f71d8702170826b2a3f5e6d7be583246c5b2eacc846fdcca6054e9ecf` |

## Where it comes from

- Model: [PaddleOCR-VL 1.6](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) by PaddlePaddle (Baidu),
  released under the Apache License 2.0.
- GGUF conversions: community repos
  [SanjeevSOLANKI/PaddleOCR-VL-1.6-GGUF](https://huggingface.co/SanjeevSOLANKI/PaddleOCR-VL-1.6-GGUF) and
  [EstevaoNaval/PaddleOCR-VL-1.6-GGUF](https://huggingface.co/EstevaoNaval/PaddleOCR-VL-1.6-GGUF).
- Change made here: the vision projector was quantized from full precision to Q4_K_M with `llama-quantize`
  (llama.cpp build 11803) to make the download smaller. The language model file is unchanged.

## Use

```
llama-server -m PaddleOCR-VL-1.6-Q5_K_M.gguf --mmproj PaddleOCR-VL-1.6-mmproj-Q4_K_M.gguf -c 4096 --jinja
```

Send a page image with the prompt `OCR:`.

## License

Apache License 2.0, same as the original model. See `LICENSE`.