What's New
German-OCR 2B Support
- New lightweight 2B model (1.5 GB) optimized for CPU/Edge
- Full llama.cpp backend integration
- ~23 tok/s on CPU, ~127 tok/s on GPU
Installation
# Standard
pip install german-ocr
# With llama.cpp support
pip install german-ocr[llamacpp]
Usage
from german_ocr import GermanOCR
# Auto-detect backend
ocr = GermanOCR()
# Or specify llama.cpp
ocr = GermanOCR(backend="llamacpp")
text = ocr.extract("invoice.png")Hardware Support
- NVIDIA (CUDA)
- Apple Silicon (Metal)
- AMD/Intel (Vulkan)
- Intel NPU (OpenVINO)
- CPU (everywhere)
Links
