Private Zero-GPU HF Space: Hebrew OCR Comparator
Upload one document image and run only the models you check.
The benchmark is built for Hebrew documents containing printed + handwritten text.
Unchecked models will be skipped and not executed.
All OCR inference is executed locally in this Space on ZeroGPU (no external inference API calls).
By default, CPU fallback is disabled so models that cannot stay on CUDA fail with clear errors.
To allow CPU fallback (slower, for compatibility only), set ALLOW_CPU_FALLBACK=1.
Tesseract is CPU-only and is disabled by default. Enable it only when you need the baseline check.