Page images stay local — model weights download on opt-in (size disclosed above the gate)
LightOnOCR
1B-class OCR VLM — deliberately not shipped (download wall).
Blocked Not shipped yet — this route is an honest placeholder so the suite hub stays complete. The badge says blocked when we know why (size / Hub maturity), not “coming soon” as marketing vapor.
- Approx download: ~0.7–0.8 GB (q4); fp16/fp32 multi-GB
- Runtime: Transformers.js + WebGPU (onnx-community/LightOnOCR-2-1B-ONNX)
- Why waiting: Blocked because q4 alone is ~0.75 GB and overlaps Granite-Docling’s layout/export story at similar download pain — shipping both would be a weighty double-gig. Try Granite instead.
Try a ready engine instead: Granite-Docling