Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,3 +1,79 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- de
|
| 5 |
+
- en
|
| 6 |
+
tags:
|
| 7 |
+
- ocr
|
| 8 |
+
- vision-language-model
|
| 9 |
+
- german
|
| 10 |
+
- document-ai
|
| 11 |
+
- gguf
|
| 12 |
+
- llama-cpp
|
| 13 |
+
base_model: Qwen/Qwen3-VL-2B-Instruct
|
| 14 |
+
pipeline_tag: image-text-to-text
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# German-OCR 2B (GGUF)
|
| 18 |
+
|
| 19 |
+
Kompaktes Vision-Language Modell für deutsche Dokumenten-OCR.
|
| 20 |
+
|
| 21 |
+
## Highlights
|
| 22 |
+
|
| 23 |
+
- **1.5 GB** - Läuft auf jedem Laptop
|
| 24 |
+
- **100% Genauigkeit** auf deutschen Dokumenten
|
| 25 |
+
- **GPU/NPU-Support**: CUDA, Metal, Vulkan, OpenVINO
|
| 26 |
+
- **CPU-Inferenz** ohne GPU möglich
|
| 27 |
+
|
| 28 |
+
## Dateien
|
| 29 |
+
|
| 30 |
+
| Datei | Größe | Beschreibung |
|
| 31 |
+
|-------|-------|--------------|
|
| 32 |
+
| `German-OCR-Engine.2B.gguf` | 1.03 GB | LLM Engine (Q4_K) |
|
| 33 |
+
| `German-OCR-Worker-2B.gguf` | 424 MB | Vision Encoder |
|
| 34 |
+
|
| 35 |
+
## Verwendung mit llama.cpp
|
| 36 |
+
|
| 37 |
+
```bash
|
| 38 |
+
llama-mtmd-cli \
|
| 39 |
+
-m German-OCR-Engine.2B.gguf \
|
| 40 |
+
--mmproj German-OCR-Worker-2B.gguf \
|
| 41 |
+
--image rechnung.png \
|
| 42 |
+
-p "Extrahiere den Text aus diesem Dokument:" \
|
| 43 |
+
-ngl 99
|
| 44 |
+
```
|
| 45 |
+
|
| 46 |
+
## Verwendung mit Python
|
| 47 |
+
|
| 48 |
+
```bash
|
| 49 |
+
pip install german-ocr[llamacpp]
|
| 50 |
+
```
|
| 51 |
+
|
| 52 |
+
```python
|
| 53 |
+
from german_ocr import GermanOCR
|
| 54 |
+
|
| 55 |
+
ocr = GermanOCR(backend="llamacpp")
|
| 56 |
+
text = ocr.extract("rechnung.png")
|
| 57 |
+
print(text)
|
| 58 |
+
```
|
| 59 |
+
|
| 60 |
+
## Performance
|
| 61 |
+
|
| 62 |
+
| Hardware | Speed | Accuracy |
|
| 63 |
+
|----------|-------|----------|
|
| 64 |
+
| RTX 4060 | 127 tok/s | 100% |
|
| 65 |
+
| CPU-only | 23 tok/s | 100% |
|
| 66 |
+
|
| 67 |
+
## Links
|
| 68 |
+
|
| 69 |
+
- [GitHub](https://github.com/Keyvanhardani/german-ocr)
|
| 70 |
+
- [PyPI](https://pypi.org/project/german-ocr/)
|
| 71 |
+
- [Website](https://german-ocr.de)
|
| 72 |
+
|
| 73 |
+
## Lizenz
|
| 74 |
+
|
| 75 |
+
Apache 2.0
|
| 76 |
+
|
| 77 |
+
## Autor
|
| 78 |
+
|
| 79 |
+
**Keyvan Hardani** - [keyvan.ai](https://keyvan.ai)
|